About dockertag
dockertag audits container image references and supplied OCI labels without contacting registries or pulling images. It is for maintainers and CI workflows that check tag pinning, repository naming, metadata, and MCP Registry server-name labels before publishing.
Key features
- 01Parse image references
- 02Check tag and digest pinning
- 03Flag uppercase repository names
- 04Check source and license labels
- 05Validate MCP server-name labels
Use cases
- 01Auditing image references before publishing
- 02Checking OCI metadata in CI
- 03Verifying MCP Registry server-name labels
- 04Reviewing image audit findings and counts
MCP support
Provides an MCP server with the audit_image_tags tool, which checks image references and supplied OCI labels, including the optional expected MCP server-name label.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
1Saved tool list · updated 1w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| images | array | Yes | Image references to audit (at least one) |
| expected_server_name | string | No | Expected value of the io.modelcontextprotocol.server.name label (optional) |
Details
From the project READMEdockertag
An image reference looks pinned until you read it closely.
repo with no tag at all silently means repo:latest;
latest itself moves; a capital letter in the repository
breaks pushes to registries that require lowercase; and a
published image can be missing the labels that say where
its source lives, what license it carries, and — for MCP
servers — the server-name label the MCP Registry checks
before it will list the image. dockertag takes the middle
road: you hand it the references and the labels you found,
and it hands back an audit.
Every reference is parsed the same careful way. The digest
is everything after the last @. A tag exists only
when the final path segment contains a : — so the colon
in localhost:5000 is a port, never a tag — and the tag is
the text after the last : in that segment. The registry
is the first segment when it contains a . or a : or is
exactly localhost; otherwise the registry is docker.io
and the whole path is the repository.
Against the parsed reference and the supplied labels, nine checks run in a fixed order:
- INVALID_REFERENCE (error) — the name part is empty, the repository is empty, or a present tag is empty. No further checks run for that image and its parsed fields are null.
- INVALID_DIGEST (error) — a digest that is not
sha256:followed by 64 lowercase hex characters. - UPPERCASE_REPOSITORY (error) — an uppercase letter anywhere in the repository.
- UNPINNED (error) — no tag and no digest, so the
reference resolves to the mutable
latesttag. - LATEST_TAG (warning) — the tag is exactly
latestand no digest pins it. - MISSING_SOURCE_LABEL (warning) — no
org.opencontainers.image.sourcelabel. - MISSING_LICENSE_LABEL (warning) — no
org.opencontainers.image.licenseslabel. - MISSING_SERVER_NAME_LABEL (error) — only checked
when you supply
expected_server_name: theio.modelcontextprotocol.server.namelabel is absent. - SERVER_NAME_MISMATCH (error) — that label is present but differs from the expected value.
The result carries each image's parsed registry, repository, tag, and digest with its finding codes, the flat list of findings (by image input order, then check order) with exact messages, and counts of images, errors, warnings, and clean images.
dockertag never contacts a registry and never pulls an
image: the labels are the ones you supply, which makes the
audit usable in CI before anything is pushed. Malformed
input — a missing or empty image list, a missing, empty,
or duplicated reference, labels that are not a
string-to-string map, a bad expected_server_name — is a
validation error naming the offending path (for example
images[1].labels.custom must be a string), and no
partial audit is produced.
CLI
npx @reneturcios/dockertag audit --input args.json
npx @reneturcios/dockertag --version
args.json contains the full arguments object:
{
"expected_server_name": "io.github.CloudCorpRecords/dockertag",
"images": [
{
"reference": "ghcr.io/cloudcorprecords/dockertag:0.1.0",
"labels": {
"org.opencontainers.image.source": "https://github.com/CloudCorpRecords/dockertag",
"org.opencontainers.image.licenses": "MIT",
"io.modelcontextprotocol.server.name": "io.github.CloudCorpRecords/dockertag"
}
},
{ "reference": "myteam/worker" }
]
}
The audit is printed as JSON:
{
"images": [
{
"reference": "ghcr.io/cloudcorprecords/dockertag:0.1.0",
"registry": "ghcr.io",
"repository": "cloudcorprecords/dockertag",
"tag": "0.1.0",
"digest": null,
"findings": []
},
{
"reference": "myteam/worker",
"registry": "docker.io",
"repository": "myteam/worker",
"tag": null,
"digest": null,
"findings": ["UNPINNED", "MISSING_SOURCE_LABEL", "MISSING_LICENSE_LABEL", "MISSING_SERVER_NAME_LABEL"]
}
],
"findings": [
{
"reference": "myteam/worker",
"code": "UNPINNED",
"severity": "error",
"message": "no tag or digest: the reference resolves to the mutable latest tag"
}
],
"counts": { "images": 2, "errors": 2, "warnings": 2, "clean": 1 }
}
(The flat findings list above is abbreviated; it holds
one entry per finding.) Exit codes: 0 on success, 1
on a runtime error (including an unreadable or invalid
input file, or invalid input), 2 on a usage error.
MCP server
npx -y @reneturcios/dockertag mcp
MCP client config:
{
"mcpServers": {
"dockertag": {
"command": "npx",
"args": ["-y", "@reneturcios/dockertag", "mcp"]
}
}
}
Tools
-
audit_image_tags— Audit container image references and OCI labels: tag pinning, lowercase repositories, source and license labels, and the MCP Registry server-name label.images(array, required, at least one) — images as{reference, labels?}objects;referenceis the full image reference as written and is unique across the set, andlabelsis the image's OCI labels as a string-to-string mapexpected_server_name(string, optional, non-empty) — the expected value of theio.modelcontextprotocol.server.namelabel; the two server-name checks run only when this is supplied
Returns
{images, findings, counts}.
Docker
docker run -i ghcr.io/cloudcorprecords/dockertag
docker run --rm -v "$PWD:/w" ghcr.io/cloudcorprecords/dockertag audit --input /w/args.json
Development
npm install
npm run build # tsc -> dist/
npm test # tsx --test test/*.test.ts (pure local, no network)
npm run dev # run the CLI from source
License
MIT — Copyright (c) 2026 Rene Turcios
Comments
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.