Skills Explorer Add Skills

mcp-builder

An agent skill by anthropics, from anthropics/skills. Tags: anthropic, api, debugging, developer-tools, mcp.

What it does

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).

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add anthropics/skills --skill mcp-builder

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/anthropics/skills
cp -r skills/skills/mcp-builder ~/.claude/skills/mcp-builder

Safety box score

Grade C, moderate risk: 35 of 100 risk points, rated 2026-10-15. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill is mainly a documentation-and-evaluation bundle for building MCP servers, not a standalone destructive tool. The main risks come from the evaluator script launching a local MCP command, sending prompts/tool outputs to Anthropic or remote servers, and writing reports; the guides also include credential and cleanup examples.

Category levels, from 0 (none) to 3 (high)
CategoryLevelWhy
Code & shell execution2 (moderate)`scripts/connections.py` can launch a caller-supplied stdio process via `StdioServerParameters(command=self.command, args=self.args, env=self.env)`, so the harness can execute local commands as part of evaluation.
Network access2 (moderate)`scripts/evaluation.py` sends conversations to Anthropic with `client.messages.create(...)`, and `scripts/connections.py` can also connect to remote MCP servers over `sse_client(...)` or `streamablehttp_client(...)`.
Filesystem reach1 (low)The harness reads an evaluation XML file and can write a report anywhere specified with `args.output.write_text(report)`; the reference TypeScript guide also shows a scoped but recursive `rm -rf dist` clean step.
Secrets & credentials1 (low)The docs require `ANTHROPIC_API_KEY` and show passing credential-bearing values like `Authorization: Bearer token123` and `GITHUB_TOKEN=ghp_xxx`, though the shipped scripts do not read secret stores.
Privilege & persistence0 (none)No file requests sudo, service installation, shell startup edits, or other OS-level privilege changes.
Instruction hijack surface1 (low)`scripts/evaluation.py` hard-codes an `EVALUATION_PROMPT` that says the model `MUST` use tools and wrap output in `<summary>`, `<feedback>`, and `<response>` tags; this is strong prompt shaping but not hidden or malicious.
Irreversible actions1 (low)The package is an evaluator, so it can drive whatever tools the connected MCP server exposes, but it does not itself publish, pay, deploy, or delete anything.
Transparency1 (low)The skill depends on external content and tooling such as `https://modelcontextprotocol.io/...`, GitHub raw docs, and `npx @modelcontextprotocol/inspector`, so some behavior comes from remote, changing sources.

Before installing

Rated 2026-10-15 by an automated review of 10 files from the skill's repository.

Source

Repository
anthropics/skills (all skills from this repository)
Path
skills/mcp-builder/SKILL.md
Branch
main
Updated
2026-09-19