Finch

# Fix list: Finch

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/finch, the October 2026 research run, assessed 8 October 2026. Grade BB, 71.6 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Finch: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Payments & pricing, 30 out of 100, up to 8.8 more on the total

Why it scored 30: Read with the hosted rubric. No x402, MPP or L402 (0). Starter is published at $65 a month per connection. Pro and Premier, which cover most providers and all writes, are priced through sales (15 of 20). A sandbox with mock data is free with a Dashboard signup and no card step is described. It reaches no real employer, so partial credit (15 of 20). A person signs up in a browser, and an employer's administrator must sign in through Finch Connect for every connection (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 2. Security & auth, 66 out of 100, up to 6 more on the total

Why it scored 66: An employer authorises each connection through Finch Connect, and the resulting bearer token is tied to that connection and to named products such as `directory`, `payment` and `ssn`. POST /disconnect revokes it. The token never expires and no rotation method was found (24 of 30). Product permissions separate reads from deduction writes, Starter is read-only, SSN needs Finch's approval, and the MCP server can be limited to GET operations. Its source notes that method blocking is a string match and can be evaded. No confirmation step for writes was found (14 of 20). Responses carry employer-entered text such as names and custom fields. The MCP page advises read-only tokens and care over which model receives the data, with no guidance on injected text (5 of 15). The Dashboard has an activity log of requests, and every response carries a request ID. No log export or API was found (10 of 15). SOC 2 Type II and a 2025 penetration test per the trust centre. No security.txt, bug bounty or disclosure policy was found (13 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 3. Agent ergonomics, 71 out of 100, up to 4.7 more on the total

Why it scored 71: `limit` and `offset` size directory, pay statement, documents and plans responses, and batch endpoints take many IDs in one call. No per-request field selection was found. Data Access Controls remove fields for a whole application on Pro and Premier. The MCP server exposes two tools (18 of 25). Offset pagination on five list endpoints, date ranges on payments and name, type and category filters on pay statement items. Most reads have no filters (14 of 20). Errors carry `code`, `name`, `finch_code` and `message` with a published table, and `Finch-Request-Id` is on every response. We saw this shape on a live unauthenticated request (17 of 20). No idempotency keys. Deduction writes return a `job_id` to poll, and enrolment is documented as an overwrite. The MCP `search_docs` tool sets readOnlyHint, and no annotations were found on the code tool (8 of 20). SDKs in six languages. Every request needs the version header (14 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 4. Reliability, 84 out of 100, up to 3.2 more on the total

Why it scored 84: Read with the hosted lines and scored on the REST API. status.tryfinch.com lists components for the API, Connect, the Dashboard and 26 provider integrations, with incident history (20). Four incidents between 9 July and 8 October 2026, three marked minor and one with no impact. They were elevated Connect errors for QuickBooks Payroll (9 to 27 July), delayed syncs for Gusto (27 July), delayed data syncs (10 August) and missing start dates on Justworks connections (24 September). None was an outage of the core API, so this reads as minor incidents only (20). Limits with numbers, 20 or 12 requests a minute per endpoint and application and 1,000 per five minutes per IP (15). 429 errors carry a `finch_code` naming the limit, and the docs say to wait 60 seconds and back off exponentially. No `Retry-After` header is documented and no idempotency keys were found for deduction writes (9 of 15). A Service Level Description commits to 99.90 per cent monthly uptime with credits, for customers whose order form includes it, which the pricing page puts on Premier (10). The REST API carries no beta label. Benefits and Documents endpoints and the MCP server are labelled beta (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 5. Transparency & trust, 74 out of 100, up to 2.3 more on the total

Made of editorial 65, provenance 82.

Why it scored 74: Closed service. The Master Service Agreement (7 August 2026), Terms of Service (24 March 2022), DPA and Service Level Description are public, and the SDKs and MCP server are Apache-2.0 (18 of 30). The privacy policy (31 July 2025) says Finch processes end-user data only on the developer customer's behalf, and the DPA commits to return or deletion at termination. Retention is described as for as long as the account is open, with no periods. The MSA lets Finch use data about the service, including data derived from customer data, to improve its services and disclose it in aggregate or de-identified form (20 of 30). A versioning page defines which changes are backward compatible and says breaking changes get a new dated version. No notice period was found, and the terms allow discontinuing any part of the service at Finch's discretion (9 of 20). The trust centre names six sub-processors with AWS us-west-2 as the hosting location, and the DPA promises 10 days' notice of additions (18 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: tryfinch.com, registered 2020-03-30 (6 years) (11 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 1 clause that costs points (7.1 of 10)
- Privacy policy: read, states 6 of the 8 things a reader expects (8.5 of 10)
- security.txt: not found (0 of 10)

## 6. Schema & documentation, 87 out of 100, up to 2.1 more on the total

Why it scored 87: One OpenAPI 3.1 file at developer.tryfinch.com/openapi.json with 55 operations and 144 schemas (25). llms.txt indexes 127 lines of guides and reference, and each page is served as Markdown (10). Guides say which endpoint to use for automated and assisted connections and how to read a 202. Operation descriptions in the specification are one line on most operations, and eight sandbox operations have none (13 of 20). Schemas carry 116 enums and required fields. POST /forward takes a free-form request for the provider (12 of 15). Curl and JSON examples in the guides, an error schema and a table of error types. The specification documents an error response on 30 of 55 operations (12 of 15). Dated versions through the `Finch-API-Version` header, and a public changelog with 344 dated entries (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 7. Maintenance & community, 84 out of 100, up to 1.4 more on the total

Why it scored 84: The changelog's latest entry is 30 September 2026, and the Python SDK 2.8.3 was published on 5 October 2026 (30). Nineteen dated changelog entries since 10 July, and ten Node SDK releases in the same period (20). Closed service with a dated changelog and support by email and help centre, plus Slack on Pro and Premier. We could not read the SDK repositories' issue replies (11 of 15). Official SDKs in six languages are current. The official MCP registry has no Finch entry (15). The SDK repositories run CI, release through release-please and have lock files. We could not confirm the CI result (8 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: live API behaviour beyond two unauthenticated requests. We had no sandbox credentials, so limits, 202 handling and writes are as documented
- unchecked: the pricing page's plan comparison table. Its cells render as icons our reader could not tell apart, so plan contents are taken from the plan cards and the docs
- unchecked: whether Starter production credentials are issued without contacting Finch. The Dashboard sits behind a login
- unchecked: the sub-processor list linked from the DPA, a Notion page our reader could not render. The six names come from the trust centre
- unchecked: the SOC 2 report and penetration test, which the trust centre releases on request
- unchecked: GitHub stars, open issues and CI results for the SDK repositories. The GitHub API refused our requests for a rate limit
- unchecked: whether 429 responses carry a `Retry-After` header in practice. None is documented
- The lead called Finch a way to reach an employer's payroll with an official MCP server. Both hold, but the MCP server is beta and local, and Finch cannot run a payroll, so `payroll.run` is not claimed

## Weaknesses

- Finch cannot create, preview or approve a payroll. Writes are limited to deductions and contributions, on Pro and Premier plans
- An access token does not expire until the connection is disconnected, and no rotation method was found in the reviewed documentation
- No idempotency keys were found, and no `Retry-After` header is documented on 429 responses
- Data is synced, not live. Automated integrations refresh every 24 hours and assisted integrations every 7 days
- Starter is $65 a month per connection for 24 providers and 15 connections. The other 250+ providers and deduction writes have no public price
- The MCP server is labelled beta, runs locally over stdio and is not in the official MCP registry

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Send `Finch-API-Version: 2020-09-17` and `Authorization: Bearer <access_token>` on every request. A request without the version header fails with 400
- Call GET /introspect first to see the connection's products, status and whether it is automated or assisted before choosing endpoints
- Treat a 202 response as data not ready, not as data. Retry with backoff, and expect up to 14 days for an assisted connection's first sync
- Stay under 20 requests a minute on organisation endpoints and 12 on pay endpoints across all tokens. Batch IDs, and on 429 wait 60 seconds
- After a deduction write, poll the job by `job_id` until it completes before enrolling individuals. Enrolment overwrites an existing enrolment
- Start the MCP server with `--code-allow-http-gets` or a token without the benefits product when only reads are intended

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.