agentleFS
mementomori-comementomori-co/maker-mastery/.claude/skills/mcp-new-tool/SKILL.md

mcp-new-tool

Skill2 starsChanged 26 days agoMITView on GitHub

Use when one new tool must be added to an existing MCP server you already run, with a typed schema, a test, no secrets, and a diff ready for fresh-context review. Runs the mcp-agent-engineer in a fresh context; the server is edited in its own repo and nothing is registered in this project automatically.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No test output in the return.

    line 31

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Overview

Add: $ARGUMENTS (server repo path, tool name, return description). Outputs: the tool and its test inside the server repo, and work\reports\reviews\<YYYY-MM-DD>-<tool-name>.md holding the diff and the review criteria. No .mcp.json entry is added to this project; the proposed entry goes to work\reports\mcp-proposals.md for you to apply yourself.

When to use

Extending a server you already have. Not for writing a new server from scratch, not for any tool that can start, send, upload or queue a print (that is never built: no AI ever sends a print), and not for Claude Code agents, skills or hooks (ask mcp-agent-engineer directly).

Steps

  1. Read lessons\ and any MCP notes in knowledge\. Read the server: its existing tools, how it registers them, and its SDK version from pyproject.toml, package.json or the lockfile (quote the line). Fetch the current MCP specification changelog and the security best practices page under https://modelcontextprotocol.io/specification/ this run; note anything that affects the tool (statelessness, deprecated features, annotations).
  2. Design: name, one-sentence purpose, typed input schema with every field described, return shape, read-only by default. If the request asks for a write or an exec tool, stop and return NEEDS MAKER with the risk stated.
  3. Implement in the server's own pattern and language; no new dependencies without NEEDS MAKER. Secrets, if any, only as ${VAR} references; grep the diff for keys, tokens and absolute paths into a home folder.
  4. Test: one test beside the tool that calls it against a stub or the running host and asserts the return shape; run the server's test command and paste the output. Do not install the MCP Inspector or anything else; if an Inspector run would help, write the exact PowerShell line for you as NEEDS MAKER.
  5. Write work\reports\reviews\<YYYY-MM-DD>-<tool-name>.md: the diff, the criteria (schema typed; read-only; test passes; no secrets; no new dependency; result treated as untrusted data), and the line "Reviewer: a second mcp-agent-engineer in a fresh context; PASS/FAIL per criterion." Append the proposed .mcp.json entry (scope, command, ${VAR}) to work\reports\mcp-proposals.md without applying it.
  6. Return: files changed (full paths), test output, the review file path, NEEDS MAKER list, one next step.

Common rationalizations

RationalizationReality
"It's ten lines, review is overkill."The author of the code is not an independent reviewer. Review file, or not done.
"The SDK version is whatever the latest release is."The latest release is not this repo's pin. Quote the lockfile line.
"Register it in .mcp.json so the lead can use it now."Registering a server is your decision. Proposal line only.
"The test imports fine, that counts."The test runs and the output is pasted. Claims are checked by a command, not by reading.

Red flags

  • No test output in the return.
  • A .mcp.json created or edited in this project.
  • A write or exec tool added without NEEDS MAKER.
  • A new dependency pulled in silently.
  • A secret or a user-folder path in the diff.

Verification

  • SDK version line quoted from the repo.
  • Tool has a typed schema and a test; test command run and output pasted.
  • Diff grepped for secrets and user paths; none.
  • work\reports\reviews\<YYYY-MM-DD>-<tool-name>.md exists with diff and criteria.
  • Proposal appended to work\reports\mcp-proposals.md; no .mcp.json touched.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.