Skip to content
anavem.com
Newsletter

Claude Skill

MCP Builder skill (Anthropic)

Anthropic skill with a four-phase guide for building Model Context Protocol servers in TypeScript or Python, with reference documents and an evaluation script that tests a server with Claude.

Maintainer
Anthropic, PBC (anthropics/skills repository)
Licence
Apache-2.0. LICENSE.txt in the skill folder is the Apache License, Version 2.0 text, "Copyright 2026 Anthropic, PBC.". SKILL.md frontmatter: "Complete terms in LICENSE.txt".
Last release
No tagged release shown (the repository releases list is empty on 2026-10-03). Latest commit on main: 2026-09-29.
Last verified Jump to what it can access ↓

What it can access

A Claude Skill is a set of instructions that may also include scripts, so check what it asks Claude to do and what those scripts can reach. This is what the listing states, based on the sources below. Anavem does not rate it safe or unsafe: check it against what you plan to use it for.

Permissions it asks for

  • Fetch web documentation: SKILL.md tells Claude to use WebFetch on the MCP specification pages and on the TypeScript and Python SDK READMEs
  • Write and build server code in the working directory (SKILL.md mentions npm run build and python -m py_compile)
  • Run MCP Inspector with npx @modelcontextprotocol/inspector to test a server
  • Run scripts/evaluation.py, which uses the anthropic and mcp Python packages (scripts/requirements.txt: anthropic>=0.39.0, mcp>=1.1.0) and connects to an MCP server over stdio (it runs the command you give it), SSE or HTTP

Data it can reach

Docs state: skills are "folders of instructions, scripts, and resources" (repository README); the mcp-builder folder holds SKILL.md, reference documents and scripts. The bundled evaluation script (scripts/evaluation.py) connects to an MCP server you point it at and creates an Anthropic API client to call a Claude model (the default model named in the script is claude-3-7-sonnet-20250219), so it sends the evaluation questions and tool results to that model. Any data a server you build can reach depends on the server you write. (SKILL.md, scripts/requirements.txt and scripts/evaluation.py, 2026-10-03)

How it is installed or connected

Claude Code (steps from the repository README, accessed 2026-10-03, https://github.com/anthropics/skills/blob/main/README.md):

  1. Register the repository as a plugin marketplace: /plugin marketplace add anthropics/skills
  2. Install the plugin that contains this skill: /plugin install example-skills@anthropic-agent-skills
  3. The README says to mention the skill by name afterwards, for example "Use the PDF skill to extract the form fields from path/to/some-file.pdf".

Claude Code skill folders: the Claude Code docs (https://code.claude.com/docs/en/skills, accessed 2026-10-03) list ~/.claude/skills/<skill-name>/ for personal skills and .claude/skills/<skill-name>/ for project skills.

Claude.ai: follow "Using skills in Claude" (https://support.claude.com/en/articles/12512180-using-skills-in-claude, accessed 2026-10-03). For an individual account the article says to enable "Code execution and file creation" under Settings > Capabilities and then toggle skills under Customize > Skills. For an organization account it points to Organization settings > Plugins & skills (Policy tab). Which accounts can use skills is stated on that article and is not asserted here.

Limitations

  • Repository README: provided "for demonstration and educational purposes only"; behavior in Claude may differ from the repository version
  • The API skills guide lists pre-built API skill IDs pptx, xlsx, docx and pdf only; this skill is not in that list, so on the API it would be a custom skill
  • The evaluation script defaults to the model claude-3-7-sonnet-20250219, which you may want to override with its --model option
  • SKILL.md tells Claude to fetch the MCP specification and the TypeScript and Python SDK READMEs from the web (WebFetch), so the workflow assumes web fetch access
  • The Claude Help Center article says: "Only install skills only from trusted sources." For a skill from a less-trusted source it says to review it before enabling, starting with the bundled files and paying particular attention to code dependencies and bundled resources like images or scripts.

Not sure what to look for? Read what to check before installing.

Quick answers

Who maintains this Claude Skill?
Anthropic, PBC (anthropics/skills repository).
What can it access?
It asks for: Fetch web documentation: SKILL.md tells Claude to use WebFetch on the MCP specification pages and on the TypeScript and Python SDK READMEs, Write and build server code in the working directory (SKILL.md mentions npm run build and python -m py_compile), Run MCP Inspector with npx @modelcontextprotocol/inspector to test a server, Run scripts/evaluation.py, which uses the anthropic and mcp Python packages (scripts/requirements.txt: anthropic>=0.39.0, mcp>=1.1.0) and connects to an MCP server over stdio (it runs the command you give it), SSE or HTTP. Docs state: skills are "folders of instructions, scripts, and resources" (repository README); the mcp-builder folder holds SKILL.md, reference documents and scripts. The bundled evaluation script (scripts/evaluation.py) connects to an MCP server you point it at and creates an Anthropic API client to call a Claude model (the default model named in the script is claude-3-7-sonnet-20250219), so it sends the evaluation questions and tool results to that model. Any data a server you build can reach depends on the server you write. (SKILL.md, scripts/requirements.txt and scripts/evaluation.py, 2026-10-03). We do not label anything safe or unsafe; read the official sources before you install.
What licence does it use?
Apache-2.0. LICENSE.txt in the skill folder is the Apache License, Version 2.0 text, "Copyright 2026 Anthropic, PBC.". SKILL.md frontmatter: "Complete terms in LICENSE.txt". Check the terms if you plan to use it commercially.
What are its limitations?
Repository README: provided "for demonstration and educational purposes only"; behavior in Claude may differ from the repository version. The API skills guide lists pre-built API skill IDs pptx, xlsx, docx and pdf only; this skill is not in that list, so on the API it would be a custom skill. The evaluation script defaults to the model claude-3-7-sonnet-20250219, which you may want to override with its --model option. SKILL.md tells Claude to fetch the MCP specification and the TypeScript and Python SDK READMEs from the web (WebFetch), so the workflow assumes web fetch access. The Claude Help Center article says: "Only install skills only from trusted sources." For a skill from a less-trusted source it says to review it before enabling, starting with the bundled files and paying particular attention to code dependencies and bundled resources like images or scripts.
When was it last released?
No tagged release shown (the repository releases list is empty on 2026-10-03). Latest commit on main: 2026-09-29. Verified Oct 3, 2026.

Sources

Related listings

Other listings in the same category.

All Claude Skills →
Claude Skill

Last30Days skill

MIT-licensed research skill that gathers recent discussion and engagement signals from social, community, prediction-market, GitHub and web sources.

By Matt Van Horn (mvanhorn on GitHub) · Verified Oct 3, 2026

Claude Skill

Discernment Nudge skill (Anthropic)

Anthropic skill that can append two or three targeted follow-up questions to help users check facts, assumptions and missing context before acting.

By Anthropic, PBC (anthropics/skills repository) · Verified Oct 3, 2026

Claude Skill

Skill Creator skill (Anthropic)

Anthropic skill that guides drafting a new skill, running test prompts against it, reviewing results in a browser viewer, benchmarking, optimizing the skill description and packaging a .skill file.

By Anthropic, PBC (anthropics/skills repository) · Verified Oct 3, 2026