Payhawk API + MCP

# Fix list: Payhawk API + MCP

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/payhawk, the October 2026 research run, assessed 8 October 2026. Grade C, 57.1 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 Payhawk API + MCP: 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, 13 out of 100, up to 10.9 more on the total

Why it scored 13: Read with the hosted rubric. No x402, MPP or L402 (0). Pricing is by quote for each module, with one published plan, the Growth programme at 149 pounds a month for single-entity firms in the UK or EEA with under 20 employees. The API page says the API is free for all Payhawk accounts (8). The Growth programme has an optional 7-day trial for eligible customers, and a development sandbox is requested through a form. Neither is open to anyone without contact (5). A person signs up as a customer, creates the key in the portal or signs in to the MCP connector (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. Reliability, 58 out of 100, up to 8.4 more on the total

Why it scored 58: Read with the hosted lines and scored on the Developer API, with the MCP server as a second surface. status.payhawk.com, a PagerDuty status page created on 22 October 2025, lists five services, one of them the Developer API (20). The page is drawn by script and its feeds returned the same shell, so the incident history went unread (5). The limit of 15 requests a second is stated on the API page, in the help centre and on every operation (15). Every operation declares a 429 with Retry-After and RateLimit-Remaining headers, and an unauthenticated call returned ratelimit headers. An Idempotency-Key header is mentioned only for creating an expense and is not declared as a parameter (10). The API page claims 99.9 per cent uptime, while section 16.3.3 of the terms disclaims service levels and no figure is published, so this is a claim and not an SLA (0). API v3 carries no beta label. The MCP docs refer to a beta period and the server is not yet in the Claude or ChatGPT directories (8).

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.

## 3. Agent ergonomics, 52 out of 100, up to 7.8 more on the total

Why it scored 52: The API has no field selection. Lists take `$take` up to 10,000 and two operations take `$include`. The MCP server has 82 tools, the lowest band, with the list filtered by the user's role and long analyses run as background jobs (12). `$skip` and `$take` appear on 7 operations, `$filter` on 11 and `$orderBy` on 2 of 75 GET operations. Paging is by offset, and the help centre and the definition disagree on the page limit (14). Errors are a code and a message, and many 400 responses are described case by case in the definition. No list of codes was found, and the 401 we received had an empty message (11). An Idempotency-Key is accepted when creating an expense other than a per diem, and PUT operations replace whole lists. MCP writes show a preview and wait for confirmation. Tool annotations could not be listed without a customer sign-in (9). Few parameters are required beyond the account ID. No official SDK was found (6).

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. Security & auth, 71 out of 100, up to 5.1 more on the total

Why it scored 71: API keys are created as read-only or full access, can be regenerated, travel in the `X-Payhawk-ApiKey` or Authorization header, and are managed by administrators or a custom role. Two system keys per account cannot be deleted by users. The MCP server uses OAuth 2 authorisation code with PKCE (S256), refresh tokens and a revocation endpoint, with no scopes published (24). Read-only keys exist. Each MCP call runs with the signed-in user's role, every write is previewed and waits for confirmation according to the docs, and payments and transfers are excluded from the MCP server (16). Supplier names, invoice text and comments reach the model, and no guidance on prompt injection was found. The MCP docs say card numbers are cut to four digits and personal details are removed from lists (4). An activity tab records changes to expenses and requests and the MCP server has a `list_audit_logs` tool. No log of calls per API key was found (9). security.txt is valid and points to a disclosure policy with no paid bounty. The trust page links SOC 1 and SOC 2 Type 2, ISO 27001, PCI DSS Level 1 and penetration test documents (18).

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.

## 5. Schema & documentation, 71 out of 100, up to 4.7 more on the total

Why it scored 71: A public OpenAPI 3.1 definition at api.payhawk.com/api/v3/docs.json with 105 paths and 179 operations, shown in a Swagger UI at developers.payhawk.com (25). payhawk.com/llms.txt and the help centre's index link Markdown copies of the help articles, the API overview and the MCP pages among them. The developer portal has no llms.txt and the reference exists only as the definition (8). 173 of 179 operations carry a description but only 34 run past 80 characters. Those that do state preconditions, such as which expense states allow an update. The MCP tool reference gives when to use and when not for five tools and names only for the rest (11). The definition has 167 enums, 18 patterns, required lists and closed objects. Filters are JSON passed as a query string, typed by a schema, and no header parameter is declared (12). 400, 401, 403 and 429 are declared on nearly every operation with one error model of a code and a message. Examples are sparse, 50 in a file of 1.4 MB (9). The version is in the path. No API changelog was found, and the help articles still cite api.payhawk.io and a v2 path (6).

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.

## 6. Maintenance & community, 53 out of 100, up to 4.1 more on the total

Why it scored 53: The OpenAPI definition was last modified on 8 October 2026 by its response header and the newest release note is dated 7 October 2026 (30). The release notes have more than three dated entries since late September 2026, but they cover the product and none we read concerns the API, so partial credit (15). Closed service with public release notes, support and partner addresses and partner solution consultants. No public forum or issue tracker was found (8). No official SDK on npm or PyPI and no entry in the official MCP registry (0). Nothing is packaged to assess (0).

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.

## 7. Transparency & trust, 78 out of 100, up to 1.9 more on the total

Made of editorial 55, provenance 100.

Why it scored 78: Closed service with public terms, last updated on 9 December 2025, which name the contracting entities and list developer APIs among the software services (15). The privacy policy (29 June 2026) and the Data Processing Addendum (26 June 2026) agree on roles. The addendum sets deletion at 90 days after termination and bars using customer personal data, prompts included, to train general models. The privacy policy gives no retention periods, and the addendum lists SOC 1 Type 1 where the trust page says Type 2 (21). No deprecation policy for the API was found. Fields are marked deprecated in descriptions without dates, and the terms give two business days' notice of scheduled maintenance (4). The addendum's schedule lists Payhawk EOOD in Belgium and Germany and two optional sub-processors, with 30 days' notice of changes. The trust page names AWS and Google Cloud and storage in Belgium and Frankfurt, but the schedule does not list the hosting providers (15).

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.

## 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: the incident history on status.payhawk.com, which is drawn by script. Its feeds returned the same page shell
- unchecked: MCP tool input schemas and readOnlyHint or destructiveHint annotations, which need a customer sign-in to list
- unchecked: the SOC, ISO 27001 and penetration test PDFs linked from the trust page were not opened
- unchecked: the fee schedules and the partner terms linked from the legal index
- Whether the MCP server enforces confirmation of writes itself or relies on the assistant. The docs say every write waits for confirmation
- Whether api.payhawk.io, cited in the help centre and the API page FAQ, still answers alongside api.payhawk.com
- Whether the Growth programme includes Developer API access and whether its 7-day trial needs a card
- The page limit. The help centre gives 1,000 records a page and the definition a maximum of 10,000
- Whether an API changelog exists behind sign-in. None is linked from the developer portal or the help centre
- The lead was right on the header, the OpenAPI 3.1 definition and the 105 paths. It did not mention the MCP server at mcp.payhawk.com, launched with the Fall '26 edition in September 2026
- No search for security incidents was made beyond the vendor's own pages. The trust page states no material breaches in the last 12 months

## Weaknesses

- No self-serve route. The API needs a Payhawk customer account, and a development sandbox is requested through a form
- No official SDK on npm or PyPI, no API changelog and no deprecation policy were found. Release notes cover the product only
- The API page's FAQ contradicts the definition on the host name, the page size and whether expenses can be created or reviewed
- The status page is drawn by script, so its incident history could not be read. The terms disclaim service levels
- The MCP server has 82 tools, is not in the official MCP registry, and no guidance on prompt injection was found

## 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 the key in `X-Payhawk-ApiKey` or as a Bearer token to `https://api.payhawk.com/api/v3`. A read-only key returns 403 on writes
- Page lists with `$skip` and `$take`. The help centre gives 1,000 a page and the definition a maximum of 10,000, so count what comes back
- Pass `$filter` as URL-encoded JSON, for example `{"status":{"$equal":"draft"}}`. Date filters compare the date part only
- Keep to 15 requests a second and wait for `Retry-After` on a 429. Do not send `Idempotency-Key` when creating a per diem expense, which returns 400
- Use a group-level key and the `/groups/{groupId}` paths for master data in a multi-entity group. Expenses and payments stay on account paths

## 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.