Skip to main content

Connect an AI agent to CheckFox over MCP and let it work through an audit.

Goal

Point an MCP client at CheckFox so an agent can read an audit, follow the official methodology criterion by criterion, and write back verdicts you then review.

Preconditions

  • The workspace owning the audit has MCP: included from Pro, an add-on on Free and Starter. The check happens per audit, so this has to be the audit's own workspace.
  • The audit already has its samples. An agent evaluates criteria; it does not decide the sampling. See methodology / sampling.
  • You are the person the agent acts as. Its rights are yours: it can never reach an audit you cannot reach, nor write to one where you can only read.

Connecting

Two ways, depending on the client.

With the native connector, no key

  1. In your MCP client, add a custom connector.
  2. Name it CheckFox and give https://api.checkfox.eu/mcp as the URL. Leave the OAuth client id and secret empty.
  3. Add it. CheckFox opens a sign-in and a consent screen.
  4. Sign in, choose whether the agent may only read or also record verdicts, and approve.

With an API key

  1. Open User settings > API keys and generate a key. Tick mcp:read so the agent can read audits and methodologies, and mcp:write only if it should record verdicts.
  2. Copy the key. It is shown once, and it is the security boundary, so treat it as a password.
  3. Configure your client to call https://api.checkfox.eu/mcp with the key as a bearer token. The setup snippets for the common clients are on checkfox.eu/mcp.

How the agent then works

One sample at a time, one criterion at a time:

  1. get_next_criterion returns the next untested criterion with its official tests, and can claim it so parallel agents do not collide.
  2. The agent decides how to verify it: reading the source, driving a browser, or running the scanner for an extra signal.
  3. It gathers evidence against every test of the criterion.
  4. It decides a status, with a user-impact level when the criterion fails.
  5. update_criterion writes the status, the findings and the suggested fix, and it moves on.

Verify success

  • The audit's progress advances, and the new entries are marked as machine-written.
  • Criteria you had already decided are unchanged. A write only lands on "Not tested".
  • Where the agent could not conclude, the criterion reads Needs review rather than Success.

Review the result

An agent's pass is a first pass, not a delivered audit. Read what it wrote before the report or the statement is built: check the ambiguous criteria yourself, rewrite the problem and solution text in your own words, and add the evidence images it cannot capture. See methodology / finding quality.

Common pitfalls

  • Granting mcp:write when read would do. Give the key only what the agent needs.
  • Expecting it to sample for you. Add the samples first, and decide the scope yourself.
  • Running it on a workspace without MCP. The tools refuse per audit, with an explicit message, even when the key itself is valid.
  • Publishing straight from an agent pass. Every verdict still needs a human reading.

Related

Last reviewed: 2026-09-03