mcp-builder (Anthropic): a Claude skill for building MCP servers

Anthropic's mcp-builder skill guides Claude through building an MCP server for an API, in TypeScript or Python, from plan to tests.

Installs: 135.5K; on skills.sh
Stars: 186.4K; on GitHub
Files: 10; besides SKILL.md
Updated: Oct 2026; last repo commit

mcp-builder is Anthropic's skill for writing MCP servers, the connectors that let an AI assistant use a service. It makes Claude work in four phases: research and plan, build, review and test, then write 10 evaluation questions. It recommends TypeScript with the current MCP SDK (v2) and also covers Python with FastMCP.

It loads when you ask Claude to build an MCP server for an API or service. It bundles reference guides, two Python scripts for running evaluations, and tells Claude to fetch the MCP spec and SDK READMEs from the web as it goes, so Claude needs web access.

Good for

Developers who want an MCP server for a service or API
Teams wrapping an internal tool so AI assistants can use it
Anyone who wants a test set that checks the server answers real questions

Install mcp-builder

Claude Code

Copy this prompt.

Install the mcp-builder skill in Claude Code. Run this command:

npx skills add https://github.com/anthropics/skills --skill mcp-builder

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).

Sections in SKILL.md

Overview
🚀 High-Level Workflow
Phase 1: Deep Research and Planning
Phase 2: Implementation
Phase 3: Review and Test
Phase 4: Create Evaluations
📚 Documentation Library
Core MCP Documentation (Load First)
SDK Documentation (Load During Phase 1/2)
Language-Specific Implementation Guides (Load During Phase 2)
Evaluation Guide (Load During Phase 4)

Key facts

Author: Anthropic
Repo: anthropics/skills
Skill folder: skills/mcp-builder
skills.sh installs: 135,476
Repo stars: 186,362
License: Terms in LICENSE.txt in the skill folder (SKILL.md does not name a license)
Last repo commit: 2026-10-28
Bundled files: 10 files besides SKILL.md
skills.sh audits: Gen Agent Trust Hub: pass; Socket: pass; Snyk: warn

What it includes

Read the MCP spec and SDK docs, then plan which tools to build
Claude studies the MCP specification and the SDK for your language, reads the API you want to wrap, and lists the endpoints to cover first. It favors broad API coverage when unsure.
Network · Phase 1: Research and planning
Set up the project, shared helpers and each tool
Builds an API client, error handling, response formatting and pagination, then each tool with input schemas (Zod or Pydantic), clear descriptions and read-only or destructive hints.
Write · Phase 2: Implementation
Build, check the code and test with MCP Inspector
Runs the build for TypeScript or a syntax check for Python, then tests the server with the MCP Inspector.
Runs code · Phase 3: Review and test
Write 10 realistic, read-only questions with checkable answers
Produces an XML file of 10 question and answer pairs that test whether an AI can use your server to answer real questions.
Runs code · Phase 4: Create evaluations
Best practices, TypeScript v2 and v1, Python and evaluation guides
Five reference files in reference/ that Claude loads as needed, plus scripts/ with connections.py, evaluation.py and an example evaluation file.
Read · Reference guides

From the official docs

Permissions

Medium risk

Claude writes code in your project and runs build and test commands, so a careless request can change files you care about. Work in a fresh folder or a git branch and read the server code before you connect it to a real account.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It runs build and test commands for you
Phase 3 builds the project and starts the MCP Inspector through npx. Approve the commands you recognize and skip the rest.
Your server can reach a real service
Testing calls the API you are wrapping. Use a test account or read-only credentials while you build.

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only switch. The skill asks for read-only operations only in the evaluation step.
Talks to	modelcontextprotocol.io and the SDK repos on GitHub (docs fetched by Claude). The API you are wrapping, when you test the server
Sign-in and keys	No account or key is named in SKILL.md.
Scopes	Writes a new server project in your working folder Runs build, syntax-check and MCP Inspector commands Fetches the MCP spec and SDK READMEs from the web

Questions

What does mcp-builder do?

It gives Claude a four-phase process for building an MCP server: plan, build, test, then write 10 evaluation questions. It picks TypeScript with the v2 SDK by default and supports Python with FastMCP.

How do I install mcp-builder in Claude Code, Claude or Codex?

Use the install box at the top of the page and pick Claude Code, Claude or Codex. Claude Code and Codex read skills from a folder on your machine. In the Claude app you upload the skill folder as a zip.

Does mcp-builder need an API key?

The skill itself does not name any key. The server you build may need credentials for the service it wraps. SKILL.md tells Claude to review that API's authentication requirements while planning.

Should I build in TypeScript or Python?

The skill recommends TypeScript with the v2 SDK for new servers and Streamable HTTP for remote ones, with stdio for local ones. Python with FastMCP is fully covered by its own guide. If your project already uses the v1 SDK, the skill tells you once that v2 is current and continues with v1 unless you ask to migrate.

Is mcp-builder safe to use?

It writes code into your project and runs build and test commands such as the MCP Inspector, and it fetches docs from the web. Read the generated server before you point it at a real account. The evaluation questions are meant to use read-only operations.

How is this different from asking Claude to write an MCP server?

The skill adds a fixed four-phase process, a best-practices guide, language guides and an evaluation step. It also tells Claude to fetch the current MCP spec and SDK READMEs before building.

Information

updated Nov 2, 2026

Repository: anthropics/skills
Folder: skills/mcp-builder

anthropics/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 2

mcp-builder SKILL.md · github.com
mcp-builder on skills.sh · skills.sh