The numbers on the comparison matrix come from a structured audit. Here is the rubric, the scoring scale, and how to read the results.
Every unified API platform is assessed against the same 17 dimensions across five groups: agent / AI ergonomics, developer experience, documentation quality, buyer access, and coverage. Verified dimensions receive a 1-to-5 score backed by dated evidence and a justification. Dimensions without sufficient evidence remain Not verified, including Apideck.
llms.txt or AI-friendly index
Does <docs>/llms.txt resolve? Is there an llms-full.txt? Is it complete?
1 = absent. 3 = present but partial. 5 = complete llms.txt + llms-full.txt covering every endpoint.
MCP server availability
Hosted, self-hosted, or none?
1 = none. 2 = repo without releases. 3 = self-hosted Docker. 5 = hosted endpoint.
Tool / action discovery
Can an agent enumerate every action from one URL?
1 = scrape required. 5 = single endpoint enumerates all tools.
First-call discoverability
How easily can a developer find a first-call example and its setup requirements?
1 = sales call required. 3 = signup + multi-step. 5 = example and required setup visible on landing. This measures discoverability, not elapsed time to a successful API call.
Sandbox accessibility
Self-serve sandbox without sales contact?
1 = none. 2 = sales-gated. 4 = free trial. 5 = self-serve sandbox.
OpenAPI spec quality
Is there a public direct URL to a complete OpenAPI spec?
1 = none. 3 = via Postman only. 5 = public URL, GitHub-published, drives the SDKs.
SDK breadth and quality
Official SDKs, language coverage, codegen vs hand-written.
1 = no SDK. 3 = 1-2 languages. 5 = 5+ languages, official, codegen-validated.
Interactive try-it
In-browser API explorer?
1 = none. 3 = Postman link only. 5 = embedded interactive explorer.
Webhook setup
HMAC docs, retry, replay, virtual webhooks for non-supporting vendors.
1 = polling only. 3 = native webhooks. 5 = native + virtual + replay + dead-letter.
Error response quality
Structured JSON, request IDs, custom HTTP codes for upstream errors.
1 = generic 4xx. 3 = structured JSON. 5 = structured + custom codes + metadata.
Search quality
Does docs search return useful results for "create an invoice"?
1 = broken. 3 = adequate. 5 = excellent semantic search.
Code examples per endpoint
Multi-language, idiomatic, present everywhere?
1 = none. 3 = curl only. 5 = multi-language, idiomatic, per endpoint.
Changelog visibility
Public, dated, breaking-change tagged?
1 = absent. 3 = present. 5 = dated + categorized + breaking-change tagged.
Migration / version handling
Explicit API versions, migration guides for breaking changes.
1 = none. 3 = versions exist. 5 = URL/header versioning + migration guides.
Free production access
Can a developer hit production endpoints without a sales call? A free production tier shortens time-to-first-call from weeks to minutes.
1 = no free option (quote-only). 2 = limited sandbox or short trial. 3 = trial with production access. 5 = perpetually-free production tier with real connections.
Connector breadth
How many integrations are pre-built and supported? Counted from the vendor’s own integrations directory, not marketing claims.
1 = under 30. 2 = 30–75. 3 = 75–150. 4 = 150–250. 5 = 250+ pre-built, normalized connectors.
Connector depth
For each connector, how complete is the unified data model? CRUD parity, resource breadth, and write coverage matter more than raw count.
1 = passthrough only, no normalized models. 2 = thin normalization with frequent gaps or read-mostly. 3 = moderate CRUD across most connectors. 4 = full CRUD on most connectors with deep resource coverage. 5 = full CRUD with write parity and deep resources across every connector.
Select any cell to inspect its evidence record. Not verified means the evidence is incomplete; it is not a score of 1. Every rating was re-audited against public sources on 15 October 2026; a dimension stays Not verified until a source supports a score.
We average only verified dimension scores, rounded to one decimal, and show the verified count alongside the result. Not verified is excluded, never counted as zero or one. With no verified dimensions there is no average. Partial averages cover different evidence and must not be treated as a vendor ranking.
First-call discoverability is an editorial assessment of finding an example. Measured time starts when the participant receives the task and ends when an authenticated response is checked against the intended account, connector and expected result. Include setup, authorization, retries and waiting; disclose any pre-provisioned credentials. A visible cURL snippet or MCP handshake is not completion.
No comparable timed first-call study is published in this scorecard yet. Report developer and agent runs separately, with task, documentation revision, client/model, sample size, success/failure counts, time budget and evidence. Agent runs also report measured tokens, tool calls and cost. These outcomes do not enter the capability average.
Verified present means evidence supports the rating. Verified absent requires explicit evidence that the capability is unavailable and receives 1. Not finding documentation is insufficient. Each verified record includes a source, explanation and review date.
developers.apideck.com/llms.txt resolves with a curated index and agent instructions, and llms-full.txt enumerates every doc page and every API operation (about 935 linked entries) with markdown links. This meets the rubric's top level: complete llms.txt plus llms-full.txt covering every endpoint.
Reviewed 2026-10-15
The vendor documents a hosted MCP server at https://mcp.apideck.dev/mcp, plus a separate hosted Docs MCP at https://developers.apideck.com/mcp. This meets the hosted-endpoint level; it does not measure authentication success, task completion or reliability. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
llms-full.txt lists every API operation from one URL, and the MCP Explorer page says each operation maps to one MCP tool. On the MCP server itself, dynamic-mode list_tools returns at most 50 tools per call and static mode currently registers only four workflow tools, so a single tools/list does not enumerate every tool. Scored 4 rather than 5 for that gap. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
The docs landing page links directly to a First API call page that lists required headers, connection prerequisites and language examples for a list-invoices read, and Get started lays out the five setup steps. Running it still needs a trial signup, a Vault connection and a consumer, so the example is one click from the landing page rather than on it (between rubric levels 3 and 5).
Reviewed 2026-10-15
Get started and the pricing page offer a self-serve 30-day free trial with full access and no sales call. Dedicated sandboxes for systems such as Workday and NetSuite are listed as an Enterprise feature. This meets the free-trial level (4).
Reviewed 2026-10-15
OpenAPI specs are published in a public GitHub repo (apideck-libraries/openapi-specs, updated 2026-10-14) and served at direct URLs such as specs.apideck.com/accounting.yml. The official TypeScript SDK README states it is built by Speakeasy, a generator that works from the OpenAPI spec. This meets the top level: public URL, GitHub-published and driving the SDKs.
Reviewed 2026-10-15
The docs list official SDKs for TypeScript/Node, Python, .NET, PHP, Java and Go, and the TypeScript SDK is generated with Speakeasy. That is six official languages from codegen, which meets the top level.
Reviewed 2026-10-15
The docs include an in-browser API Explorer covering every Unified API. It builds requests from the OpenAPI spec and runs them against a live consumer connection after sign-in. This meets the embedded interactive explorer level. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
The webhooks guide documents native webhooks, Virtual Webhooks that poll connectors without native support, HMAC-SHA256 signatures, idempotency keys and a 10-attempt retry schedule over about 28 hours. Failed events can be redelivered only by contacting support, and there is no self-serve replay or dead-letter queue, so it falls short of level 5. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
The errors page documents one JSON envelope (status_code, error, stable type_name, message, detail, ref link) and a list of typed errors by status. Connector errors pass through the downstream HTTP status with the downstream response in detail, and proxy responses carry x-apideck-request-id and x-apideck-downstream-error headers. This meets the top level: structured, custom codes and metadata.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
A public API changelog lists 920 dated entries tagged by API, change type, connectors and resources, sourced from the openapi-specs CHANGELOG. Breaking changes are called out in entry text and the versioning policy refers to a breaking-changes section, but the change-type taxonomy shown (fix, coverage-expansion, other) has no breaking tag. Scored 4.
Reviewed 2026-10-15
The versioning page states there is one live schema with no version headers or per-account pinning. It documents a six-month deprecation runway, a table of what counts as breaking, migration paths, and machine-readable x-apideck-sunsetting markers in the spec. Strong policy, but no URL or header versioning, so it does not reach level 5; scored at the middle level.
Reviewed 2026-10-15
Pricing offers a 30-day free trial with full access, then paid Launch and Scale plans from $599/month. There is no perpetually free tier. This meets the trial-with-production-access level (3).
Reviewed 2026-10-15
The integrations directory links 180 unique connector pages across Accounting (60), HRIS (59), CRM (22), Ecommerce (17), ATS (10), Issue Tracking (9), File Storage (5) and Auth Only (4). The headline says 200+. Both counts fall in the 150 to 250 band (level 4).
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
docs.chift.eu serves llms.txt (with per-section indexes, including an API reference index of 158 endpoint pages) and a 1.4 MB llms-full.txt. About 18 non-deprecated operations in the published OpenAPI spec are missing from the index, so it falls just short of rubric level 5 (every endpoint).
Reviewed 2026-10-15
Chift documents a hosted remote MCP server at https://mcp.chift.eu/mcp with OAuth 2.0 and dynamic client registration, alongside an open-source local server. This meets the rubric's hosted level (5). It does not measure authentication success, task completion or reliability, and the evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
Docs say the hosted MCP tools mirror the Unified API one-to-one across all verticals and can be listed from one endpoint. Each session lists only the tools for the scopes granted to a single consumer, and there is no public tool catalog, so it sits below rubric level 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
A public OpenAPI 3.1 spec is available at a direct URL (156 paths). No evidence was found that it is GitHub-published or generates the SDKs, so it sits below rubric level 5.
Reviewed 2026-10-15
The SDKs page lists two official SDKs, Python and Node.js, plus the OpenAPI file for Postman. Two languages is rubric level 3 (1-2 languages).
Reviewed 2026-10-15
API reference endpoint pages include an embedded Try it playground. An authenticated API Explorer inside the platform can also execute requests. That meets rubric level 5 (embedded interactive explorer).
Reviewed 2026-10-15
Native webhooks are configurable by UI or API and signed with HMAC (X-Chift-Signature). Events cover connections, flows, issues and data layer changes. No retry, replay or dead-letter behavior is documented, so it meets rubric level 3 (native webhooks).
Reviewed 2026-10-15
Errors are structured JSON with status, message, error_code and an optional detail field. A catalog of custom error codes covers connector-side failures (for example ERROR_CONNECTOR_AUTHENTICATION and ERROR_CONNECTOR_MAINTENANCE). No request ID is documented in the error shape, so it sits below rubric level 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Endpoint reference pages carry generated request snippets in several languages: cURL, Python requests, PHP, Go and Ruby. These are generic HTTP-client snippets, not SDK-idiomatic code, so it sits below rubric level 5.
Reviewed 2026-10-15
Public changelog pages are rolled up by month (latest July 2026) under a Product heading. Entries have no individual dates and no breaking-change tags, so it meets rubric level 3 (present).
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The integrations catalogue in Chift's llms.txt lists 143 integration pages; 13 are marked Coming soon, leaving about 130 live. That is in the 75-150 band (rubric level 3).
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
https://docs.codat.io/llms.txt and llms-full.txt both return 404. The marketing site serves an llms.txt, but it lists only recent blog content and a sitemap link, with no documentation or endpoint coverage. That falls short of the rubric's partial level (3).
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The First steps guide lists an account and API key as prerequisites and says to get in touch to discuss creating an account. The homepage's Get Started and Book a Meeting buttons both lead to a meeting booking. That meets the rubric's sales-call level (1).
Reviewed 2026-10-15
Codat documents a Sandbox integration with mock accounting, commerce and banking data, but its First steps guide says to get in touch before an account can be created. The sandbox is therefore sales-gated (rubric level 2).
Reviewed 2026-10-15
Codat publishes JSON and YAML OpenAPI specs for each API on GitHub, and its libraries page says the official SDKs are generated from that OpenAPI specification. That meets rubric level 5: a public URL, published on GitHub, that drives the SDKs.
Reviewed 2026-10-15
The libraries page lists official, OpenAPI-generated SDKs in three languages (TypeScript, Python, C#); PHP and Ruby libraries are community-maintained. Three official languages fall between rubric levels 3 (1-2 languages) and 5 (5+).
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Codat's webhook service documents signed webhooks, automatic retries on a retry schedule, delivery logging, event replay and mock events, and a 2024 update added replay of missed messages from the past two weeks. No dead-letter queue is documented, so it sits below rubric level 5.
Reviewed 2026-10-15
Errors return structured JSON with statusCode, service, error, correlationId and validation details. The documented status codes are standard HTTP, with no custom codes for upstream platform errors, so it sits between rubric levels 3 and 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The public Updates log has dated posts (latest October 1, 2026) tagged by product and area, with dedicated deprecation and action-required tags. That meets rubric level 5: dated, categorized and tagged for breaking changes.
Reviewed 2026-10-15
Separate product API versions are published (for example, Sync for Payables v1 and current, plus Bank Feeds v2), and a change policy promises at least three months' notice for breaking changes. All specs share the unversioned base https://api.codat.io with no URL or header version selector, so it meets rubric level 3 (versions exist).
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The All integrations page says 30+ platforms and lists about 36 entries across accounting, banking, commerce and bank feeds. That is in the rubric's 30-75 band (level 2).
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
docs.conductor.is/llms.txt resolves and indexes every guide and all API reference endpoints, and llms-full.txt (about 400 KB) includes the endpoint pages such as Create an invoice and List all invoices. This meets the 5 level: complete llms.txt plus llms-full.txt covering every endpoint.
Reviewed 2026-10-15
The vendor documents a hosted remote MCP server at https://mcp.conductor.is/ authenticated with a secret key, with setup for Claude Code, Cursor and Codex. This satisfies the hosted-availability rubric; it does not measure authentication success, task completion or reliability. Evidence is vendor documentation, not an independent runtime test.
Reviewed 2026-10-15
A single OpenAPI 3.1 file (about 88,000 lines, 262 operations) enumerates the whole API, llms.txt lists every endpoint, and the MCP server exposes the entire API through search_docs and execute tools. This meets the 5 level: one URL enumerates all actions.
Reviewed 2026-10-15
The quickstart lays out self-serve signup, a testing project, end-user creation, the auth flow, an API key and SDK code, but requires your own QuickBooks Desktop test instance on Windows. This meets the 3 level: signup plus multi-step setup.
Reviewed 2026-10-15
Signup is self-serve with a testing project, and connections to QuickBooks sample company files or trial installations are free, but Conductor provides no hosted sandbox: developers must build their own QuickBooks Desktop test instance (guide uses AWS WorkSpaces). This sits between the 2 (sales-gated) and 4 (free trial) levels.
Reviewed 2026-10-15
The OpenAPI 3.1 spec is downloadable from the conductor-is/quickbooks-desktop-node GitHub repository, which is the Stainless-generated Node SDK the spec drives. This meets the 5 level: public URL, GitHub-published, drives the SDKs.
Reviewed 2026-10-15
Conductor documents two official, fully typed SDKs, Node.js (conductor-node) and Python (conductor-py), generated with Stainless. This meets the 3 level: 1-2 languages.
Reviewed 2026-10-15
API reference pages, for example List all invoices, render an embedded Try it playground in the docs. This meets the 5 level: embedded interactive explorer.
Reviewed 2026-10-15
The FAQ states that Conductor does not support webhooks because QuickBooks Desktop is a local application, and recommends user-initiated sync or scheduled polling instead. This meets the 1 level: polling only.
Reviewed 2026-10-15
Every error returns a consistent object with type, custom code (e.g. QBD_CONNECTION_ERROR, INTEGRATION_CONNECTION_NOT_ACTIVE), integrationCode passthrough, httpStatusCode, userFacingMessage and requestId, and each request carries a Conductor-Request-Id header. This meets the 5 level: structured plus custom codes plus metadata.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The published OpenAPI spec attaches JavaScript and Python SDK code samples (x-codeSamples) to each of its 262 operations, and these render on the reference pages. This meets the 5 level: multi-language, idiomatic, per endpoint.
Reviewed 2026-10-15
Product updates are published with per-day dates (183 entries, latest 2026-10-09) and each item is prefixed with its area such as API. Breaking changes are not consistently tagged, so it falls short of the 5 level.
Reviewed 2026-10-15
Requests use a URL-versioned base (api.conductor.is/v1) and a migration guide covers moving from the old conductor-node package to QuickBooks Desktop API v2. The API v2 naming does not match the /v1 URL and there is no header versioning, so it falls just short of the 5 level.
Reviewed 2026-10-15
Pricing is $49 per month per in-use QuickBooks Desktop connection; connections to sample company files or trial installations are free, and free test exemptions for production files require contacting support. This meets the 2 level: limited free sandbox use, no free production tier.
Reviewed 2026-10-15
Conductor's API covers a single integration, QuickBooks Desktop. This meets the 1 level: under 30 connectors.
Reviewed 2026-10-15
For its one connector Conductor documents 256 endpoints and 20,000+ fields with create, update, list and delete across QuickBooks Desktop types, while listing some missing types (DataExt custom fields, ListDel, ItemFixedAsset) and no batch writes. This meets the 4 level: full CRUD with deep resource coverage.
Reviewed 2026-10-15
developer.tryfinch.com/llms.txt lists 119 pages including every API reference endpoint, and llms-full.txt (about 380 KB) contains the full endpoint pages. Meets the 5 level.
Reviewed 2026-10-15
Finch documents a beta MCP server distributed as the npm package @tryfinch/finch-api-mcp, run locally through npx with a connection access token. No hosted endpoint is documented, so this is a self-hosted server at the 3 level.
Reviewed 2026-10-15
A public OpenAPI 3.1 document at developer.tryfinch.com/openapi.json enumerates the API (44 paths), and llms.txt lists every endpoint page. One URL lists every endpoint, but as an API spec rather than an agent tool listing, so this sits just below 5.
Reviewed 2026-10-15
The Quickstart shows the first curl calls, but they need a self-serve sandbox signup first, then a redirect URI, a Connect session, sandbox login, and a code-for-token exchange before the first data call. Meets the 3 level: signup plus multi-step.
Reviewed 2026-10-15
Developers sign up for a free sandbox account in the dashboard without contacting sales. Every account includes a Finch Sandbox (100 mock connections) and free Provider Sandboxes (5 live-provider connections). Meets the 5 level: self-serve sandbox.
Reviewed 2026-10-15
Finch's llms.txt links a public direct OpenAPI 3.1 spec at developer.tryfinch.com/openapi.json. The SDKs are generated with Stainless, but the public file is not shown to be their source, so this stays below the 5 level.
Reviewed 2026-10-15
Finch lists six official backend SDKs (Node, Python, Java, Kotlin, Go, Ruby), generated with Stainless, plus JavaScript and React Connect SDKs. Meets the 5 level: 5+ official codegen languages.
Reviewed 2026-10-15
API reference pages, for example Directory, include a Try it button that opens an in-browser request playground. Meets the 5 level: embedded interactive explorer.
Reviewed 2026-10-15
Finch sends native account, job and per-record data change webhooks with HMAC-SHA256 signatures (Finch-Signature) and retries failed deliveries with exponential backoff. Events are dropped once retries are exhausted, so there is no replay or dead-letter queue. Meets 4.
Reviewed 2026-10-15
Errors return a documented JSON schema (code, name, finch_code, message) with a catalog that separates upstream cases such as upstream_provider_error (502) and upstream_rate_limit_exceeded (429). No error metadata object is documented, so this meets 4.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Endpoint reference pages show request examples in cURL plus SDK-idiomatic Node, Python, Java, Kotlin, Go and Ruby snippets, for example on Directory. Meets the 5 level: multi-language, idiomatic, per endpoint.
Reviewed 2026-10-15
The public changelog is linked from the docs. Entries are individually dated (latest October 13, 2026) and tagged by category (API / Platform, Bug / Fix, Integration Update, New Feature, Finch Connect, Sandbox), but there is no breaking-change tag. Meets 4.
Reviewed 2026-10-15
Every request must send a dated Finch-API-Version header (current 2020-09-17), and Finch documents what it treats as backward-compatible on a dedicated API Changes page. No migration guide between versions was found, so this is header versioning short of the 5 level.
Reviewed 2026-10-15
The cheapest plan, Starter, costs $65 per connection per month (24 providers, read-only); Pro and Premier are contact-sales. Free use is limited to the sandbox. Meets the 2 level: limited sandbox, no free production tier.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Organization, Payroll, Benefits and Documents are read products, and writes are limited to Deductions (and pay statement rules); the Starter plan is read-only. Most providers are assisted integrations refreshed every 7 days rather than automated. Meets the 2 level: read-mostly normalization.
Reviewed 2026-10-15
docs.kombo.dev/llms.txt resolves and links per-section indexes (691 developer documentation pages, plus HRIS and ATS indexes), and llms-full.txt (about 3.7 MB, roughly 750 pages) includes the endpoint reference pages. Meets the 5 level: complete llms.txt plus llms-full.txt covering the endpoints.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
A single public OpenAPI 3.1 document at api.kombo.dev/openapi.json enumerates 109 paths across HRIS, ATS, LMS, Assessment and AI Apply. This is one URL listing every endpoint, but it is an API spec rather than an agent tool listing, so it sits just below the 5 level.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Kombo links its official OpenAPI specification at a public direct URL (api.kombo.dev/openapi.json, OpenAPI 3.1). Its Speakeasy-generated SDKs build from a separate auth-gated internal spec (api.kombo.dev/internal/sdk.openapi.yaml), so the public file is not shown to drive the SDKs; that keeps it below the 5 level.
Reviewed 2026-10-15
Kombo publishes three official server-side SDKs (TypeScript, Python, Ruby), generated with Speakeasy, plus a frontend Connect SDK. Three codegen languages sit between the 1-2 language level (3) and the 5+ language level (5).
Reviewed 2026-10-15
Endpoint reference pages have a Try It button that opens an in-browser playground for testing endpoints, for example Get employees. Meets the 5 level: embedded interactive explorer.
Reviewed 2026-10-15
Kombo sends native data-changed webhooks signed with HMAC-SHA256 (X-Kombo-Signature) and subscribes to upstream webhooks where tools support them; where they do not, data-changed fires after the next scheduled sync. The page does not show replay or a dead-letter queue, so this meets 4 rather than 5.
Reviewed 2026-10-15
Errors use a documented JSON envelope with a programmatic code (for example PLATFORM.INPUT_INVALID), title, message and a log_url pointing to the dashboard log, and upstream rejections are categorized under Kombo codes. Kombo tells clients not to rely on HTTP status codes and documents no custom HTTP codes, so this meets 4.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Endpoint pages show generated request snippets in several languages (cURL, Python, JavaScript), and the Try It panel produces code for more. The snippets are generic HTTP rather than idiomatic SDK calls on every endpoint, so this meets 4.
Reviewed 2026-10-15
changelog.kombo.dev is public and linked from the docs index. Entries are dated (latest October 11, 2026) and labeled by product (General, ATS, HRIS, LMS, Assessment, SFTP, DATEV), with deprecations called out in prose but no breaking-change tag. Meets 4.
Reviewed 2026-10-15
The API is versioned in the URL (api.kombo.dev/v1, with EU, US and CA hosts) and deprecated endpoints are marked in the reference. The docs index lists no migration guides between API versions. Meets the 3 level: versions exist.
Reviewed 2026-10-15
Pricing is quote-based (annual platform fee plus per-customer fee) and every plan CTA is Get a Demo. The FAQ says free testing is a proof-of-concept phase with sandbox access before a commercial decision. Meets the 2 level: limited sandbox or trial, no perpetual free production tier.
Reviewed 2026-10-15
Kombo's connector directories list about 209 unique tools across HRIS (101 pages), ATS (122), LMS (11) and Assessment (20), excluding the Kombo sandbox connectors. That falls in the 150-250 band, score 4.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
docs.maesn.com/llms.txt resolves and indexes every guide, integration page and API reference endpoint, and llms-full.txt (about 2 MB) includes the full page content. This meets the 5 level: complete llms.txt plus llms-full.txt covering every endpoint.
Reviewed 2026-10-15
Maesn's MCP docs tell users to clone the public maesndev/maesn-mcp GitHub repository and run it locally with Node; the repository has no published releases and no hosted endpoint is documented. This meets the 2 level: repo without releases.
Reviewed 2026-10-15
llms.txt links every endpoint page and the OpenAPI files at docs.maesn.com/openapi.json (101 paths) and openapi-v2.json (4 v2 paths), so an agent can enumerate the API from one index URL. The surface is split across two spec files and the MCP server lists its tools only in the repo README, so it falls short of a single tool-enumerating endpoint (5).
Reviewed 2026-10-15
The quick start shows a first-call curl example and lists the setup steps, but step 1 says to reach out to Maesn to obtain an API key, and every pricing tier is Book a demo. The example is discoverable, but setup requires vendor contact, between the 1 (sales call required) and 3 (signup + multi-step) levels.
Reviewed 2026-10-15
The Sandboxes page says Maesn provides sandboxes for every supported system, but Maesn-owned sandboxes with test data require contacting Maesn; the pricing FAQ describes a proof-of-concept phase with sandbox access. This meets the 2 level: sales-gated.
Reviewed 2026-10-15
A public OpenAPI 3.0 spec is served at docs.maesn.com/openapi.json and linked from llms.txt, and the Speakeasy-generated TypeScript SDK repository on GitHub publishes the spec it is generated from (.speakeasy/out.openapi.yaml). This meets the 5 level: public URL, GitHub-published, drives the SDK.
Reviewed 2026-10-15
Maesn documents one official SDK, @maesn/typescript-sdk, marked beta and generated with Speakeasy. This meets the 3 level: 1-2 languages.
Reviewed 2026-10-15
API reference pages, for example Get invoices, render an embedded Try it playground button in the docs, alongside a public Postman workspace. This meets the 5 level: embedded interactive explorer.
Reviewed 2026-10-15
Maesn offers unified webhook subscriptions via POST /webhooks for nine systems, with HMAC-SHA256 signatures in X-MAESN-SIGNATURE; other systems are documented as polling. No retry, replay, dead-letter or virtual webhooks are documented. This meets the 3 level: native webhooks.
Reviewed 2026-10-15
Errors return a structured errors object with statusCode, a standardized type (e.g. target_system_error, resource_not_supported), context metadata (targetSystem, unifiedApi, timestamp), validationErrors and passthrough downstreamErrors. No request ID is documented, so it sits between the 3 (structured JSON) and 5 (structured + custom codes + metadata) levels.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The public changelog is dated by month and grouped into New integration, New endpoints, Improvements and Changes sections, with Beta badges. Breaking changes are not explicitly tagged, so it falls short of the 5 level.
Reviewed 2026-10-15
The API exposes v1 and v2 surfaces (v2 paths such as /accounting/v2/contacts in openapi-v2.json), and the changelog notes v1 behavior aligned to v2. No migration guide was found in the docs index. This meets the 3 level: versions exist.
Reviewed 2026-10-15
The pricing page lists Start, Scale and Enterprise tiers, all Book a demo with a quote after a scoping call; the FAQ says Maesn can be tested for free in a proof-of-concept phase with sandbox access. This meets the 2 level: limited sandbox or trial, no free production tier.
Reviewed 2026-10-15
The docs integrations section lists 31 systems (including AbaConnect and Abacus, DATEV Rechnungswesen and DATEV Unternehmen Online as separate entries). This meets the 2 level: 30 to 75 connectors.
Reviewed 2026-10-15
The Endpoint coverage table shows uneven depth: Exact Online, Lexware Office and Business Central expose 17 to 26 resources, while FreshBooks, Pennylane and Visma eAccounting expose only customers and suppliers and several others three to five resources. This meets the 2 level: thin normalization with frequent gaps.
Reviewed 2026-10-15
docs.getmembrane.com/llms.txt resolves and lists guides plus every API reference endpoint (about 240 entries), and llms-full.txt (about 590 KB) includes the API reference pages. Meets level 5.
Reviewed 2026-10-15
The vendor documents a hosted cloud MCP server at https://api.getmembrane.com/mcp/integrate-anything with OAuth or bearer-token auth. This satisfies the hosted-availability rubric; it does not measure task completion or reliability.
Reviewed 2026-10-15
GET /actions lists actions across universal, integration and connection layers, with text search and up to 1,000 results per page. Actions are workspace elements often built on demand rather than a fixed catalog, and a connection is required for intent-based listing, so it is scored below level 5.
Reviewed 2026-10-15
The AI agent quickstart shows a one-line skill install and the hosted MCP config, and the REST API page shows a first curl call with the required token. The setup is visible publicly but still needs an account login, so it sits between levels 3 and 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
A public OpenAPI file is served at docs.getmembrane.com/api-reference/membrane-api.json and drives the API reference. The URL given on the REST API page (api.getmembrane.com/-json) returned 404 on 2026-10-15, and no GitHub or SDK-codegen link was found, so it is scored below level 5.
Reviewed 2026-10-15
The SDK page lists a TypeScript/JavaScript core SDK, React and Vue wrappers built on it, and a CLI. That is a single language, which meets level 3 (1-2 languages).
Reviewed 2026-10-15
API reference pages render an embedded Try it playground (for example List actions). Meets level 5, embedded interactive explorer.
Reviewed 2026-10-15
Events reach your app via Flows or external event subscriptions with a webhookUri. Data collection events are detected through native webhooks or configurable polling, and failed deliveries are retried with exponential backoff. Signature verification, replay and dead-letter handling for these deliveries are not described, so it falls short of level 5.
Reviewed 2026-10-15
All API errors use a documented JSON structure with type, key, message, a data object and a nested causedByError, mapped to standard HTTP status codes. There are no custom HTTP codes for upstream errors, so it is scored below level 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
A public changelog has dated entries filterable by API Contract, CLI, Platform and SDK, with Added, Changed, Fixed, Removed and Breaking sections. Meets level 5.
Reviewed 2026-10-15
The API uses X.Y versioning, where the major version changes for breaking changes, and supports pinning with an API-Version header, with API Contract versions tracked in the changelog. No dedicated migration guides were found, so it is scored below level 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
docs.merge.dev/llms.txt resolves with agent instructions, OpenAPI links and per-product section indexes (for example merge-unified/llms.txt) that list endpoint pages. llms-full.txt redirects back to llms.txt, so there is no separate full-content file. Scored 4: complete index, no llms-full.
Reviewed 2026-10-15
The vendor documents a hosted Agent Handler MCP endpoint (https://ah-api.merge.dev/api/v1/tool-packs/<id>/registered-users/<id>/mcp) and a hosted docs MCP server. This meets the hosted-endpoint level; it does not measure authentication success, task completion or reliability. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
tools/list on an Agent Handler MCP URL returns every tool in the configured Tool Pack in one response (optional pagination of 100 per page), and search_tools supports narrowing. Enumeration is scoped to a configured Tool Pack rather than the whole catalog from one public URL, so scored 4. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
The docs home links to a Unified Quickstart that lists prerequisites (free signup, API key, test Linked Account) and shows a curl GET to /api/hris/v1/employees plus an interactive tester with mock data. The example is one click from the landing page and a signup is required, between rubric levels 3 and 5.
Reviewed 2026-10-15
Signup is free and self-serve. The quickstart creates test Linked Accounts from the dashboard, and the SDK docs note that TEST_ACCOUNT_TOKEN returns dummy data. Pricing lists test Linked Accounts on every plan (3 on Launch). This meets the self-serve sandbox level.
Reviewed 2026-10-15
Public direct OpenAPI 3.1 URLs exist per category, indexed at docs.merge.dev/openapi.json (the Accounting spec fetched with 132 paths). A GitHub-published spec and spec-driven SDK generation were not confirmed, so scored below the top level.
Reviewed 2026-10-15
The SDKs page lists official all-category SDKs for Node, Java (Kotlin/JVM), Python, Go, Ruby and C#/.NET, with GitHub repos. The page says the SDKs are in various stages of active development, and codegen validation was not confirmed, so scored 4.
Reviewed 2026-10-15
Endpoint reference pages render an in-page Try it request runner (seen on Accounting List Invoices), and the quickstart embeds an interactive API tester with mock data. Postman collections are also offered. This meets the embedded interactive explorer level. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
Merge documents native webhooks (Linked Account sync, changed-data events), test POST delivery, HMAC-SHA256 signature verification via X-Merge-Webhook-Signature, and ingestion of third-party webhooks. Retry, replay and dead-letter behaviour were not documented on these pages. This meets the native-webhooks level. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
Failed writes return a structured JSON envelope of errors and warnings, each with problem_type, title, detail and source, drawn from an enumerated code list that includes PROVIDER_ERROR, RATE_LIMITED and RELINK_NEEDED. The reference covers POST endpoints, and no request ID was documented there. Scored 4: structured with machine codes. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The public changelog has dated entries with permalinks, filterable by product, category and update type (New, Improved, Fixed, Deprecated, Security). There is no explicit breaking-change tag, so scored 4.
Reviewed 2026-10-15
The versioning page says the APIs are at v1 with no major releases planned and commits to notice before any major version, breaking change or deprecation. Paths include /v1/. Versions exist, but no migration guides were found. This meets the middle level.
Reviewed 2026-10-15
The Unified Launch plan includes 3 production Linked Accounts free, with paid usage starting at $650/month for up to 10. This is a perpetually free production tier with real connections (level 5), though capped at 3 accounts.
Reviewed 2026-10-15
The per-category integration directories in the Unified docs list 221 unique integrations (231 category listings): HRIS 88, ATS 63, Ticketing 37, CRM 21, Accounting 14, File Storage 5, Knowledge Base 2, Chat 1. The vendor's llms.txt claims 260+. Scored on the directory count, which falls in the 150 to 250 band.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
nango.dev/docs/llms.txt resolves and lists every HTTP API endpoint page, and llms-full.txt (about 860 KB) includes the HTTP API reference pages; a separate api-catalog.txt indexes 1,054 providers. Meets the level 5 bar of a complete llms.txt plus llms-full.txt.
Reviewed 2026-10-15
The vendor documents a hosted Management MCP server at https://mcp.nango.dev/mcp (Streamable HTTP, OAuth), plus per-session hosted MCP URLs for agent sessions. This satisfies the hosted-availability rubric; it does not measure task completion or reliability.
Reviewed 2026-10-15
GET /integrations/{uniqueKey}/functions lists the functions and catalog actions for one integration, and an agent session MCP exposes a nango_tool_search meta tool. Discovery is per integration or search-based: the session tools/list returns only pinned and meta tools, so no single call enumerates every tool, which places it below level 5.
Reviewed 2026-10-15
The public quickstart shows the full setup (free signup, pre-created GitHub integration, test connection, API key) and a copyable Node and cURL first call, plus an agent prompt. The example and setup are visible without an account, but reaching it takes several dashboard steps, so it sits between levels 3 and 5.
Reviewed 2026-10-15
Developers can sign up for free without sales contact and add test connections in the dashboard; the pricing page lists a $0 Free plan with hard-capped monthly limits. Meets level 5, self-serve sandbox.
Reviewed 2026-10-15
The OpenAPI spec is public at nango.dev/docs/spec.yaml and published in the NangoHQ/nango GitHub repo, and it drives the API reference pages. Evidence that it generates the SDKs was not found, so it is scored below level 5.
Reviewed 2026-10-15
The only available backend SDK is Node (@nangohq/node), plus a frontend JavaScript SDK; the Python, Java, Ruby, Go and Rust SDK pages say "Coming soon. Use the REST API in the meantime." One language family meets level 3 (1-2 languages).
Reviewed 2026-10-15
The HTTP API reference pages render an embedded Try it playground (for example on Trigger an action). Meets level 5, embedded interactive explorer.
Reviewed 2026-10-15
Nango sends HMAC-SHA256 signed webhooks (X-Nango-Hmac-Sha256) with key rotation and retries non-2xx deliveries twice with exponential backoff. It also forwards external webhooks and combines webhooks with polling in real-time syncs. Replay and dead-letter handling are not described, so it falls short of level 5.
Reviewed 2026-10-15
The OpenAPI spec defines structured JSON errors with an error object holding a machine-readable code and message, and the troubleshooting guide shows nested upstream error payloads. Request IDs and custom HTTP codes were not documented in the pages reviewed, so it meets level 3 (structured JSON).
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
A public changelog in the docs has dated entries (latest October 6, 2026) and calls out deprecations with "Am I impacted?" and "What to do" sections linking migration guides. Entries are labelled by topic rather than tagged by category or breaking change, so it sits below level 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The pricing page lists a $0/month Free plan covering API and MCP auth, tools and syncs with hard-capped limits that reset monthly (10 connections, 10 compute hours, 10 GB). Meets level 5, a perpetually free production tier with real connections.
Reviewed 2026-10-15
The vendor's API catalog lists 1,054 provider slugs. That is above the 250+ threshold for level 5. These are pre-built auth integrations with templates; normalization is assessed separately under connector depth.
Reviewed 2026-10-15
Nango's own docs say unification is customer-built: "You define the model your product wants to use" and implement provider-specific functions to map to it. Pre-built per-provider tools exist, but there is no vendor-maintained normalized data model, which matches level 1 (no normalized models).
Reviewed 2026-10-15
docs.useparagon.com/llms.txt resolves and indexes the docs, API reference, ActionKit sub-indexes and OpenAPI specs, and llms-full.txt (about 2 MB) includes the API and ActionKit reference pages. Meets level 5.
Reviewed 2026-10-15
Paragon's docs describe ActionKit as "available as a self-hosted MCP" via the open-source useparagon/paragon-mcp repo, which ships a Dockerfile and docker-compose file but no tagged releases. No hosted MCP endpoint is documented. Meets level 3, self-hosted Docker.
Reviewed 2026-10-15
GET /projects/{project_id}/tools lists every Tool available to a Connected User with input JSON schemas, including user-specific custom fields. Meets level 5, single endpoint enumerates all tools.
Reviewed 2026-10-15
The ActionKit getting-started guide requires adding the Paragon SDK, activating an integration in the dashboard and connecting an account before the first API call, with a hosted demo and playground as testing alternatives. Meets level 3, signup plus multi-step.
Reviewed 2026-10-15
The pricing page offers "Start free trial" linking to self-serve signup at dashboard.useparagon.com/signup, and the docs list Trial-plan limits (for example 7 days or 50k logs retention). Meets level 4, free trial.
Reviewed 2026-10-15
The docs publish direct OpenAPI files, including /actionkit/openapi.json and /actionkit/openapi-standard.json, which drive the API reference. No GitHub publication or SDK codegen link was found, so it is scored below level 5.
Reviewed 2026-10-15
The documented SDK is the JavaScript @useparagon/connect package for the browser (npm, yarn, pnpm, bun). One language meets level 3 (1-2 languages).
Reviewed 2026-10-15
ActionKit API reference pages render an embedded Try it playground. Meets level 5, embedded interactive explorer.
Reviewed 2026-10-15
ActionKit Triggers deliver events to a configured webhook URL with a signing secret, and support both WEBHOOK and POLLING subscription models. Managed Sync webhooks are at-least-once with retries, stable delivery IDs for deduplication, and delivery logs. Replay and dead-letter queues are not described, so it falls short of level 5.
Reviewed 2026-10-15
ActionKit errors return structured JSON with message, code, status and a meta object (for example code 7203 for an expired token), and Managed Sync documents stable error codes with suggested actions. Custom HTTP status codes for upstream failures are not shown, so it sits below level 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Paragon publishes separate dated changelogs for product updates, SDK, REST API and Paragraph. The API release notes page exists specifically to list breaking changes, with affected API and migration steps. Meets level 5, dated, categorized and breaking-change tagged.
Reviewed 2026-10-15
API release notes are labelled by version (for example 2026-04, 2026-07), state what changes and what to do, and preserve old paths for backward compatibility. Pinning a version by URL or header is not described in these pages, so it is scored below level 5.
Reviewed 2026-10-15
Pricing is quote-based ("Get Custom Quote") with no free tier listed. The only free option is a trial with documented Trial-plan limits. Meets level 2, limited trial.
Reviewed 2026-10-15
The vendor's integrations catalog in the docs lists 124 integrations, and ActionKit's supported-integrations page lists 59. Meets level 3 (75 to 150).
Reviewed 2026-10-15
Normalized schemas exist only for read-side Managed Sync objects (File, CRM Contact, Company, Deal, Custom Object, Ticket). Writes go through per-integration ActionKit tools, not a unified model. Meets level 2 (thin, read-mostly normalization).
Reviewed 2026-10-15
On 2026-10-15, https://docs.rutter.com/llms.txt and llms-full.txt returned HTTP 404, as did https://www.rutter.com/llms.txt. No AI index resolves at the docs root, which is rubric level 1.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The Quickstart walks through self-serve signup, getting API keys, copying a sandbox connection access token, and then a Postman collection or cURL commands. That is signup plus several steps (rubric level 3).
Reviewed 2026-10-15
A developer account is self-serve, and the Free Starter Plan includes testing the API with sandbox data. Organizations created after December 2023 come pre-loaded with sandbox connections. That is a self-serve sandbox (rubric level 5).
Reviewed 2026-10-15
Each API version has its own public OpenAPI 3.1 spec at a direct URL; the 2024-08-31 spec has 193 paths. No evidence was found that it is GitHub-published or drives official SDKs, so it sits below rubric level 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Rutter sends webhooks for all platforms, including ones without native webhooks, by computing changes on each incremental sync. It also offers real-time webhooks for QuickBooks Online and Xero. Payloads are HMAC-SHA256 signed and retried up to 6 times with exponential backoff, after which the webhook is marked failed; no replay or dead-letter is documented, so it sits below rubric level 5.
Reviewed 2026-10-15
Errors return error_type, error_code, error_message and an error_metadata object. Custom HTTP codes 450, 451 and 452 separate upstream platform errors from Rutter's own 400, 401 and 429. That meets rubric level 5: structured, with custom codes and metadata.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
A public changelog posts weekly dated entries (latest October 8), each tagged by platform and endpoint. Breaking changes are listed per version on a separate API Upgrades page rather than tagged in the changelog, so it sits below rubric level 5.
Reviewed 2026-10-15
Clients choose a dated API version through the X-Rutter-Version header on the /versioned root URL, and webhook payloads can also be versioned. The API Upgrades page lists breaking changes for each version with upgrade steps. That meets rubric level 5: header versioning plus migration guides.
Reviewed 2026-10-15
The Free Starter Plan is 30 days, covers sandbox data, and lets you try QuickBooks, Xero, FreshBooks and Zoho Books. Live production access needs the Full Access Plan, which is quote-based. That is a short trial (rubric level 2).
Reviewed 2026-10-15
The docs Supported Platforms list has 90 entries: 83 platforms plus 7 bank-feed connectors. That is in the 75-150 band (rubric level 3). The marketing integrations directory shows a smaller set, about 56 unique platforms.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
docs.stackone.com/llms.txt indexes every docs page, including per-connector pages and the platform API reference, and llms-full.txt resolves with the full documentation content (about 4 MB). This meets the top level.
Reviewed 2026-10-15
The vendor documents a hosted remote MCP server at https://mcp.stackone.com/mcp, with OAuth setup guides for ChatGPT, Claude, Cursor, VS Code and others. This meets the hosted-endpoint level; it does not measure authentication success, task completion or reliability. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
GET https://api.stackone.com/actions returns action metadata for all connectors, with optional filters by connector, account or action key. The docs also describe tool_search and tool_execute meta-tools for search-based discovery. One endpoint can enumerate every action (level 5). Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
The docs landing page links to a Quickstart that walks through dashboard steps: create a connector profile, link an account, then run actions in the no-code Playground. The quickstart shows no code or curl first-call example, and a signup plus several steps are needed. This meets the signup-plus-multi-step level.
Reviewed 2026-10-15
Pricing offers free Starter and OEM Core plans with no credit card that include all connectors, MCP access and monthly credits, and the docs describe a self-serve Playground for running actions. This meets the self-serve level with no sales contact.
Reviewed 2026-10-15
The docs link public direct OpenAPI URLs for the Documents and IAM APIs (for example api.eu1.stackone.com/oas/iam.json, fetched). No complete platform or actions spec URL or GitHub-published spec was confirmed, so this is scored below the top level.
Reviewed 2026-10-15
The API SDKs page lists official client SDKs for TypeScript, Ruby, PHP (Beta), Java (Beta) and C# (Beta), and the docs add TypeScript and Python agent SDKs. That is five or more languages, but three client SDKs are marked beta, so scored 4.
Reviewed 2026-10-15
The Mintlify API reference pages render an in-page Try it request runner (seen on List Actions). This meets the embedded interactive explorer level. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
The webhooks page documents connector events and platform events, HMAC-SHA256 signatures in x-stackone-signature and secret rotation. Data Sync events notify changes for providers with no webhooks of their own, retrying up to 20 times. No replay or dead-letter feature was documented in these pages, so scored 4. Evidence is vendor documentation, not a runtime test.
Reviewed 2026-10-15
The error reference states that all StackOne API errors share one JSON shape (statusCode, message, timestamp, errorCode, provider_errors). It documents non-generic codes such as 412 for suspended linked accounts, 501 for features a provider does not support and 502 for upstream provider errors. This meets the top level.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The API SDKs page states that the API reference provides copyable code snippets per endpoint in JavaScript, Python, PHP, Go, Java and Ruby. Language tabs are visible on the List Actions reference page. This meets the multi-language per-endpoint level.
Reviewed 2026-10-15
Each connector has a public changelog with dated, semver-labelled releases tagged with categories including breaking, bug-fix, new-actions and data-sync, and it lists the affected actions. This meets the dated, categorized and breaking-change-tagged level.
Reviewed 2026-10-15
Connectors use semver, and each connector profile can be pinned to latest, a major or minor range, or an exact version. This lets teams test new versions before rolling out breaking changes. Explicit versions exist with pinning, but no migration guides were confirmed, so scored 4.
Reviewed 2026-10-15
The pricing FAQ says both pricing models have a free plan with no credit card that includes all 550+ pre-built connectors, 34,000+ actions, MCP access and a monthly credit allowance. This meets the perpetually-free production tier level; usage is capped by credits.
Reviewed 2026-10-15
The docs index has connector pages for about 549 distinct connectors, and pricing cites 550+ pre-built connectors. That is well above the 250+ threshold. Many are action-based connectors rather than normalized unified models; that aspect belongs to connector depth.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
docs.unified.to/llms.txt (about 120 KB, 377 links) indexes guides, MCP docs, per-category API overviews and the OpenAPI spec. llms-full.txt (about 4.8 MB) includes the endpoint tables and data models for every API category. Meets the 5 level.
Reviewed 2026-10-15
Unified.to documents a hosted Streamable HTTP MCP server at mcp-api.unified.to/mcp (US), with EU and AU endpoints. Meets the hosted-availability rubric; it does not measure authentication success, task completion or reliability. Evidence is vendor documentation, not an independent runtime test.
Reviewed 2026-10-15
A documented GET /tools endpoint on the MCP API returns every MCP tool for a connection, with filters and output formats for OpenAI, Anthropic, Gemini, Cohere, Grok and Groq. Meets the 5 level: a single endpoint enumerates all tools.
Reviewed 2026-10-15
The quick start needs four steps before the first call: sign up, copy the workspace ID and API key, activate an integration in the sandbox, and embed the Authorization component to get a connection ID. An Agent Signup endpoint can create a workspace and key for coding agents. Meets the 3 level: signup plus multi-step.
Reviewed 2026-10-15
After self-serve signup, developers can switch to a Synthetic Data (sandbox) environment that returns mock data with mocked credentials, either in the app or with env=Sandbox on the API. Meets the 5 level: self-serve sandbox.
Reviewed 2026-10-15
A public OpenAPI 3.0 spec is at api.unified.to/openapi.json, with per-category files as well, and the docs say it is updated at least daily. The official SDK Speakeasy workflow uses that URL as its source. Meets the 5 level.
Reviewed 2026-10-15
Seven official SDKs (TypeScript, Python, PHP, Java, Go, C#, Ruby) are listed, generated with Speakeasy from the public OpenAPI spec. Each shows a release dated within the last two days. Meets the 5 level.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Native and virtual webhooks are documented, the latter for integrations without native support, with signed payloads (sig, sig256). For virtual webhooks, failed deliveries get immediate retries and progressive backoff for about two weeks, resuming from the last delivered page. Native webhooks get only 3 immediate retries and no dead-letter queue is documented, so this meets 4.
Reviewed 2026-10-15
The REST reference documents status codes, including 501 for features an integration does not support. An unauthenticated request returned a structured JSON body (statusCode, error, message, attributes). No error-code catalog, request ID or upstream metadata is documented, so this meets 3.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
The public changelog has dated entries (latest 2026-10-14), each labeled by API category or as an enhancement. No breaking-change tag was found. Meets 4.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Paid plans start at $750 a month. Grow and Pro offer a 30-day free trial, Grow is marked Production Capable, and a tester account includes 2,500 free API requests. There is no perpetually free production tier. Meets the 3 level: trial with production access.
Reviewed 2026-10-15
The integrations directory lists 1,357 unique integrations across 37 API categories (338 not marked beta), and pricing claims 1,358+. Either count is well above 250, score 5.
Reviewed 2026-10-15
No public source supporting a score was recorded in the latest audit. Missing evidence does not establish that this capability is absent.
Review date: pending re-audit.
No supporting source retained for this rating.
Vendors ship new docs portals constantly. Each evidence record carries its own review date; a page update is not a re-audit. Evidence older than 90 days should be re-audited before making a procurement decision.
Found something we missed or got wrong? Email hello@apideck.com with the source URL and the dimension. We will rerun the audit on that platform and update the matrix.