Aller au contenu principal

The CheckFox MCP server, which lets an AI agent read an audit, follow the official methodology, and record verdicts.

Cette documentation est disponible en anglais uniquement.

What the MCP server is

A Model Context Protocol (MCP) server hosted by CheckFox, so an AI agent can work inside an audit. It is the read-and-write layer, not the auditor: judgement stays with the agent and with you.

Endpoint: https://api.checkfox.eu/mcp. Guide: checkfox.eu/mcp.

What an agent can do with it

Per criterion, the agent reads the official tests and methodology, chooses how to verify them (page source, a real browser to exercise keyboard, focus and reflow, or the CheckFox scanner as an extra signal), then records a status with findings and a suggested fix.

Nine tools:

ToolPurposeScope
list_auditsThe audits the caller can accessread
get_auditOne audit, its samples and its progressread
list_criteriaA sample's criteria with their current statusread
get_next_criterionThe next untested criterion, with its methodologyread
get_criterion_methodologyAny criterion's methodology, by referential and numberread
run_scannerRun the automated scanner on a sample, in light mode, dark mode or bothwrite
measure_contrastMeasure the real text contrast of chosen elements of a samplewrite
update_criterionRecord a status, findings and a fixwrite
release_criterionRelease a claim the agent holds on a criterion (for example, it decided not to audit it)write

run_scanner and measure_contrast need the scanner on the audit's workspace; each call counts as one scan of the workspace's monthly allowance.

The referential is also exposed as read-only resources, both accepting a language parameter: an index of every criterion of a standard, and any criterion's methodology on demand.

Light mode and dark mode

Many sites follow the system theme (prefers-color-scheme). run_scanner and measure_contrast take color_scheme: light (default), dark, or both. both renders the page twice in one call and tags each finding with its theme. Contrast passing in light mode can fail in dark mode, so scan both on any site with a dark theme.

Measuring contrast precisely

Automated scanners leave text over a gradient, decorative halo or image as "needs review". measure_contrast settles it. The agent names text elements by CSS selector and gets, for each:

  • a measurement from the computed styles of the text and every layer behind it, gradients included, covering every scroll position of a fixed background;
  • a measurement from the real screen pixels behind the text, with the text hidden, covering background images and filters, at one scroll position;
  • a verdict against the required ratio (4.5:1, or 3:1 for large text): pass, fail, review (the measurements disagree, or contrast varies across the text) or unknown.

Nothing is written to the audit: the result is evidence for criterion 3.2.

The CheckFox scanner (in the app, the browser extension and run_scanner) now measures text over gradients automatically: the checkfox-contrast-gradient rule.

Why it does not damage an audit

  • It never overwrites a person. Writes only land on criteria still "Not tested"; human verdicts are never touched.
  • Read and write are separate permissions. A key can have mcp:read only.
  • Roles still apply. A key of someone who can only view an audit can read it, never write.
  • Work can be claimed. get_next_criterion can atomically claim the returned criterion with a lifetime, so parallel agents do not collide. A claim can be released early.
  • Writes are marked as machine-written, with the model, like an in-app AI fix suggestion. See findings.
  • Traffic is rate limited per key.

What it doesn't do

  • It doesn't create the audit: audit creation and sampling are still done in CheckFox. See procedures / create an audit.
  • It doesn't replace the auditor's judgement.

Plan availability

Included from Pro; the MCP add-on on Free and Starter. Checked per audit, against the plan of that audit's workspace, not of the key holder. See plans and roles.

Where it lives in the UI

  • User Settings > Integrations - generate an API key with the MCP scopes.
  • My Workspaces > Integrations - whether the workspace has MCP access.

Dernière révision : 2026-11-02