v1.3.0

Changed

• A partial audit now fails min-score, as it fails the CLI's
  --fail-under. Without a working credential an auth-gated server is audited
  only on its auth, TLS and transport surface, and that score used to pass the
  gate. The job now fails with a message naming the partial audit and how to
  pass a credential. A workflow that relied on a green partial run needs a
  credential, or no min-score. min-readiness is unchanged.

Added

• Show failure messages and repair hints in report summaries, with literal text
  rendering and explicit informative/counted readiness labels. Old reports remain supported.

v1.2.0

Added

• sarif-path input: the failed rules as SARIF 2.1.0 for GitHub code
  scanning. Set it to a file and add a github/codeql-action/upload-sarif
  step, and every failed rule becomes an alert in the repository's Security
  tab, keyed on the rule and the target so a re-run updates it in place. The
  file is written before any gate runs, so a failing build still carries its
  findings. Needs mcpscore 1.14.0 or later (the default, latest, qualifies);
  an older version fails the job with a message naming the requirement.
  The README has the workflow.

v1.1.1

Fixed

• The comment token no longer reaches the audited server. The audit
  subprocess inherited the step environment, including the github-token
  input, so a local target (server code from the pull request under review)
  could read a write-capable token. The action now strips its own token from
  the environment it launches mcpscore with. Secrets you export yourself,
  such as MCPSCORE_TOKEN, are still passed through on purpose.
• The readiness line in the comment no longer claims readiness is never
  part of the main score. Since mcpscore 1.1.0 a server on the modern
  lifecycle has its readiness points counted in the score, and the report
  says so (readiness.counted_in_main). The comment and job summary now
  read "counted in the main score" or "informative, not counted in the main
  score" accordingly.

Changed

• The audit runs with PYTHONFAULTHANDLER=1, so a native crash in the CLI
  prints a traceback in the job log instead of a bare exit code.

v1.1.0

Added

• Picks up a mcpscore.toml from the checkout. Rules turned off or
  re-ranked in the repository's configuration (or [tool.mcpscore] in
  pyproject.toml) shape the score the action gates on, and the PR comment
  names the file, what it changed, and the rules that tripped a configured
  [gate]. Needs mcpscore 1.12.0 or newer; older engines behave as before.
• A release script and a publish workflow. make release runs the
  preflight (main, clean, in sync, changelog section, tag free, CI green)
  and creates the GitHub Release; the Publish workflow then moves the
  floating major tag to it in CI, where the move is auditable. Previously
  the major tag was moved by hand.
• make all mirrors CI (ruff, pyright, pytest, zizmor), like the other
  mcpscore repositories.

Changed

• A CLI gate exit now publishes the report before failing the job.
  mcpscore exits 3 for --fail-under and a configured [gate] and 4 for a
  --smoke failure, after writing its JSON report. The action used to treat
  any non-zero exit as "could not audit" and skip the report, outputs, job
  summary, and comment; it now publishes them and then fails the job with
  the named reason.

v1.0.1 - Runtime refresh

Bumps the embedded astral-sh/setup-uv from 8.2.0 to 10.0.1 (transient manifest-timeout resilience, current uv checksums) and refreshes the version-input example. The audited mcpscore engine is unaffected: the action always runs the latest engine unless you pin the version input.

v1.0.0 - Audit MCP servers in CI

First release of the mcpscore GitHub Action — Lighthouse for MCP in your CI.

Audit an MCP server on every pull request, fail the build when quality drops below a threshold you set, and get the report as a PR comment.

- uses: mcp-box/mcpscore-action@v1
  with:
    target: https://your-server.example/mcp
    min-score: 80

What it does

• Score gate — fail the job when the main score falls below min-score (a percentage).
• Readiness gate — optionally require readiness for the upcoming MCP 2026-07-28 spec via min-readiness (reported separately, never mixed into the main score).
• PR comment — posts one report comment and updates it in place on re-runs: score, a pass/fail breakdown by severity, spec version and era, and every failed check by rule_id.
• Job summary + outputs — the same report in the run summary, the raw JSON on disk, and every number exposed as a step output (score, percentage, era, readiness-*, ...).
• Deterministic — no API keys, no LLM calls; the same server state always produces the same result, in seconds.

Works against remote servers (Streamable HTTP / SSE) and local .py / .js servers.

Links:

• Documentation: https://docs.mcpscore.dev/github-action
• Scoring methodology: https://docs.mcpscore.dev/methodology
• mcpscore CLI: https://github.com/mcp-box/mcpscore

Full changelog: https://github.com/mcp-box/mcpscore-action/commits/v1.0.0