mcpField
S
shipcheck

shipcheck

Local
github.com/CloudCorpRecordsDeveloper Tools0Updated 3d ago

About shipcheck

shipcheck checks whether a Node, Python, or hybrid release agrees across its applicable publishing surfaces. It is for maintainers who need a single GO or NO-GO verdict backed by per-surface evidence.

Key features

  1. 01Check tags, commits, and release metadata
  2. 02Verify npm and PyPI versions
  3. 03Check MCP Registry and GHCR status
  4. 04Skip inapplicable publishing surfaces
  5. 05Report expected and actual values

Use cases

  1. 01Verify a release before announcing it
  2. 02Find publishing discrepancies in CI
  3. 03Check a specific package version across registries

MCP support

The MCP server exposes `check_release` for full repository checks and `check_npm`, `check_pypi`, and `check_registry` for individual surfaces.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"cloudcorprecords-shipcheck": {
"command": "npx",
"args": [
"-y",
"@reneturcios/shipcheck"
]
}
}
}
Run commandnpx -y @reneturcios/shipcheck

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
npm@reneturcios/shipcheck0.2.0
ocighcr.io/cloudcorprecords/shipcheck:0.2.0

Tools

3

Saved tool list · updated 3d ago

ParameterTypeRequiredDescription
repostringYesRepo in owner/repo form, e.g. CloudCorpRecords/greenlight
skiparrayNoSurfaces or groups to skip: npm, registry, ghcr, release, github
requirearrayNoSurfaces or groups to force as hard checks: npm, registry, ghcr, release, github
versionstringNoExpected version, e.g. 0.15.1. Omit to read it from package.json on the default branch.
npmPackagestringNonpm package name when it cannot be inferred from package.json at the tag
serverNamestringNoMCP Registry server name, defaults to io.github.<Owner>/<repo>

Details

From the project README

shipcheck

Did that release actually ship? One command, every surface, one verdict.

A modern release touches many public surfaces: the git tag, the GitHub release, the package metadata in the repo, npm or PyPI, the MCP Registry, and the container image on GHCR. Each one can fail silently while the others look fine. The publish workflow goes green, and then: npm has the version stuck in a staged hold, the registry still points at the last release, the container tag never got pushed, or, the worst one, the git tag points at a commit that does not exist.

shipcheck checks every surface against one expected version and prints a single verdict: GO, or NO-GO with the exact surfaces that disagree and the evidence for each one. It speaks both ecosystems: Node repos are read from package.json and npm, Python repos from pyproject.toml / setup.cfg / setup.py and PyPI, and hybrid repos (a Python core with an npm wrapper, like wallcheck) are checked on every carrier they actually have.

Why this exists

shipcheck was not designed in the abstract. It is the manual verification gauntlet from shipping Greenlight and its sibling tools, turned into a tool. That gauntlet caught real failures:

  • A release where the git tag pointed at a nonexistent commit object. npm, the registry, and the sources were all real; only checking that the tag resolves to a real commit exposed it. That is why tag-commit is a hard check.
  • npm staged publishing holds, where a brand new package version sits in review and the registry shows only a placeholder. To a user, the release simply does not exist. That is why npm-version and npm-latest are separate checks.
  • Registry publishes dispatched before npm verified, leaving the MCP Registry pointing at a version npm did not have yet. That ordering lesson is why shipcheck treats "listed" and "latest" as different facts on every surface.

What it checks

surfacewhat it proves
git-tagTag v<version> exists on the repo
tag-commitThe tag resolves to a real commit object
tag-versionpackage.json at the tag carries the expected version
pyproject-versionEvery Python version carrier at the tag agrees: pyproject.toml ([project] or [tool.poetry]), setup.cfg, setup.py, and dynamic versions resolved from the file they point at (hatch, setuptools attr: / file:, __version__)
server-versionserver.json at the tag, its package versions and its OCI tags all agree
github-releaseA published (not draft) GitHub release exists for the tag
npm-versionThe npm package has this version published (only when the repo has an npm package)
npm-latestnpm dist-tags.latest points at this version
pypi-versionThe PyPI package has this version published with files (Python projects)
pypi-latestThe latest PyPI release is this version
registry-versionThe MCP Registry lists this version for the server
registry-latestThe MCP Registry marks this version latest and active
ghcr-tagThe GHCR image has this version tag
ghcr-latestThe GHCR image also has a latest tag
python-testsSoft: Python test files exist at the tag, with the runner shape (pytest config, or unittest shape)
dockerfileSoft: a Dockerfile exists in the tag tree
readmeSoft: a README exists at the repo root
mcp-entrySoft: an MCP server entry point exists (a *-mcp or server console script, or server.json)

Verdict rules:

  • GO: every hard check passes. Exit code 0.
  • NO-GO: any hard check fails or errors. Exit code 1, ready for CI.
  • SKIP: a surface that does not apply is skipped, never failed. A repo with no MCP Registry entry skips the registry checks; a repo with no public container image skips GHCR; a Python project that is not on PyPI skips the PyPI checks; a Python project with no package.json skips npm instead of failing it. GHCR checks are soft by default and become hard with --require ghcr.
  • Every check reports expected vs actual and the source URL. A failed fetch is reported as an error with its HTTP status, never guessed.
  • The report names the detected ecosystem (node, python, hybrid, unknown). A Python app with no metadata carriers at all (requirements files and sources only) is still detected from its tag tree, and reports its missing version carrier as a SKIP with the evidence, not as a silent pass.

Usage

CLI
# Full check. Version is read from package.json or the Python metadata
# (pyproject.toml, setup.cfg, setup.py) on the default branch.
npx @reneturcios/shipcheck check CloudCorpRecords/greenlight

# Python and hybrid repos work the same way
npx @reneturcios/shipcheck check CloudCorpRecords/wallcheck 0.5.0

# Check a specific version
npx @reneturcios/shipcheck check CloudCorpRecords/greenlight 0.15.1

# Machine readable output, custom surface names, force GHCR to be required
npx @reneturcios/shipcheck check owner/repo 1.2.0 --json \
  --npm @scope/pkg --pypi pkg --server io.github.Owner/repo --require ghcr

# Single surfaces
npx @reneturcios/shipcheck npm @reneturcios/greenlight 0.15.1
npx @reneturcios/shipcheck pypi wallcheck 0.5.0
npx @reneturcios/shipcheck registry io.github.CloudCorpRecords/greenlight 0.15.1

Example output:

shipcheck CloudCorpRecords/greenlight v0.15.1: GO
  [PASS] git-tag: Tag v0.15.1 exists (object type commit)
  [PASS] tag-commit: Tag resolves to real commit f52939005a0f
  [PASS] tag-version: package.json at v0.15.1 is version 0.15.1
  [PASS] github-release: GitHub release v0.15.1 is published
  [PASS] npm-version: @reneturcios/[email protected] is published on npm
  [PASS] npm-latest: npm latest points at 0.15.1
  [PASS] registry-version: MCP Registry lists io.github.CloudCorpRecords/greenlight 0.15.1 (status active)
  [PASS] registry-latest: MCP Registry latest is 0.15.1 and active
  [PASS] ghcr-tag: GHCR image has tag 0.15.1
  [PASS] ghcr-latest: GHCR image has a latest tag
GO: every hard surface agrees on this version.

Inference, so the common case needs one argument: the npm package name is read from package.json at the tag or on the default branch, the PyPI package name from the Python metadata, the registry server name from server.json when present (else io.github.<Owner>/<repo>), and the GHCR image defaults to ghcr.io/<owner>/<repo> lowercased. Override any of them with --npm, --pypi, --server, or --ghcr. Skip whole groups with --skip npm,pypi,python,server,registry,ghcr,release,github.

Set GITHUB_TOKEN for the higher authenticated GitHub API rate limit. Public surfaces need no other credentials; GHCR uses the standard anonymous token flow for public images.

MCP server

Add to your MCP client config:

{
  "mcpServers": {
    "shipcheck": {
      "command": "npx",
      "args": ["-y", "@reneturcios/shipcheck", "mcp"]
    }
  }
}

Tools:

  • check_release: the full check: repo, optional version, optional npm/PyPI/server overrides, require and skip lists.
  • check_npm: one npm package and version: published, and latest points at it.
  • check_pypi: one PyPI package and version: published with files, and it is the latest release. Absent from PyPI reports as skipped.
  • check_registry: one MCP Registry server and version: listed, and marked latest.
Docker
docker run -i ghcr.io/cloudcorprecords/shipcheck mcp
docker run --rm ghcr.io/cloudcorprecords/shipcheck check CloudCorpRecords/greenlight 0.15.1

The toolkit

shipcheck is Wave 1, package 1 of a planned family of small, single-purpose developer tools by the same author, all built the same way: one bin that serves a CLI and an MCP server, deterministic checks with evidence, no invented answers.

  • greenlight: pre-flight checks for open-source contributions: is this issue actually landable?
  • prwatch: attention report for your open PRs: what needs you right now.
  • claimcheck: stale-claim states and parked fixes on forks.
  • gatefile: extract hidden contribution gates into machine-readable contribution-gates.json.
  • pkgverdict: pre-install verdicts for npm packages: GO, RISKY, FAKE, or MISSING.

Development

npm install
npm run build
npm test   # 40 tests, all surface logic runs on injected fixtures, no network

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.