Skip to main content
Agent skillFirst-run guide

Mcp Builder

Design and implement Model Context Protocol servers so an agent can call external tools with typed schemas instead of stuffing API docs into a skill.

Get started

Get the complete skill folder, including its scripts and reference files.

·

When to use this skill

Use MCP Builder when the agent must call a live system (GitHub, Stripe, a database, a browser) with authentication and up-to-date schemas. Skills hold procedural knowledge; MCP servers hold tools.

When not to use it

If the workflow is "remember how we write release notes", that is a skill, not MCP. Do not wrap a static style guide in a server.

How to install

Where will you use it?

Install the publisher’s plugin

This installs a bundle that can include other skills. Choose this method or the manual alternative below; you only need one installation.

  1. Run these commands in Claude Code:

    Claude Code CLI
    /plugin marketplace add anthropics/skills
    /plugin install example-skills@anthropic-agent-skills
  2. Ask Claude Code to use Mcp Builder on a small task and confirm that it loads. Check the setup requirements in the publisher’s guide if it cannot run.
Alternative: install the folder manually

Use these steps if you choose the manual folder method. If you installed the plugin above, skip this alternative.

  1. Claude Code personal install: place the skill and its bundled files at ~/.claude/skills/mcp-builder/SKILL.md. For a shared project, commit the folder at .claude/skills/mcp-builder/ instead.
  2. Current Claude Code detects SKILL.md changes in watched directories during the session. If you just created a top-level skills directory, run /reload-skills. Check that the skill is listed before testing its trigger.
  3. For Claude chat or Cowork in the Desktop app, manage uploaded skills in Customize > Skills; copying files into a local Claude Code folder is not a Desktop install. Other agents may use different install locations.

Full install guide for Claude, Cursor, and Codex →

What supports this recommendation?

67/100 provisional content score. This measures documentation, practical detail and source signals. It is not a measured task success rate or a security certification.

Documentation
67/100
Practical detail
50/100
Source evidence
70/100
Source signals
94/100
Model compatibility evidence
  • claudeuntested

    No model-specific signal or recorded compatibility test was found.

  • chatgptuntested

    No model-specific signal or recorded compatibility test was found.

  • geminiuntested

    No model-specific signal or recorded compatibility test was found.

  • copilotuntested

    No model-specific signal or recorded compatibility test was found.

  • llamauntested

    No model-specific signal or recorded compatibility test was found.

  • perplexityuntested

    No model-specific signal or recorded compatibility test was found.

  • mistraluntested

    No model-specific signal or recorded compatibility test was found.

  • grokuntested

    No model-specific signal or recorded compatibility test was found.

Inferred compatibility is based on content signals. Check the app, its available tools and setup requirements before running a skill.

Source reviewedExample execution recorded

Try one small task

Best first exercise: expose one read-only tool over a tiny local fixture.

See actual outputs and execution limits →

What you need

  • A development environment for a supported MCP SDK and a compatible local MCP test client.
  • A small documented data fixture; production credentials are unnecessary for this exercise.

Build a local inventory lookup with one successful and one missing-item case.

First task prompt
Use mcp-builder to design a local read-only inventory MCP server. The fixture has SKU A001 with 7 units and B002 with 0 units. Expose one lookup tool with a typed SKU input. Return an explicit not-found result for C003. Choose an SDK, explain the transport, implement the tool, and show how to invoke all three cases with an MCP client. Do not claim the server works until those calls have run.

Expected result

A specification to check against, not output from an AI run.

A minimal local MCP server, clear startup instructions, and three invocation cases with distinguishable results.

  • A001 returns 7 and B002 returns 0; missing C003 is distinct from zero stock.
  • Invalid input is handled predictably, with no production credential required.
  • A client invocation demonstrates the tool interface, not just a direct function unit test.

Where this exercise stops

  • A local fixture does not test deployment, authentication, rate limits, or a production service.
  • Static instructions alone may be sufficient when no callable external tool is needed.

Recorded example run

These files came from actual execution in a Codex session. They were produced and checked within the same session; this is scoped evidence, not independent review or a measured success rate.

Inventory MCP: actual stdio client calls

Executed · Automated Codex exercise

Codex agent; Node.js 24.13.0, Python 3.12.3, Linux 7.0.0-1010-aws. Exact model version not exposed.

Implemented a local TypeScript server with the official MCP SDK, compiled it, and launched it as a separate stdio process from an SDK client. Captured the actual initialize, tools/list and tools/call messages.

Download the actual outputs

Recorded checks and observations

  • Observed: TypeScript compilation passed with SDK 1.32.0 and Zod 4.6.5.
  • Observed: Eight protocol tests passed: discovery, A001=7, B002=0, C003 not found, three invalid-input cases and an unchanged repeated read.
  • Observed: The actual handshake negotiated protocol 2025-11-25; server stderr was empty in the saved transcript.
Scope, limitations and pinned sources
  • The agent read the pinned upstream instructions and used their prescribed tools. Native Claude skill loading and general cross-agent compatibility were not tested.
  • The inventory is a synthetic local fixture; no external inventory service or production credentials were used.
  • Authentication, deployment, production rate limits, MCP Inspector UI and model tool selection were not tested.
  • This does not compare the agent with a no-skill baseline.
  • The executor was Codex in this conversation; the exact model version was not exposed.
  • These examples do not represent independent human users, customer outcomes or a compatibility certification.
Read the execution record (JSON) →
Source and evidence

The upstream instructions informed this guide. The exercise and acceptance checks are an OpxSkills test recipe. Where a run is recorded, its actual outputs, environment and limits appear separately above. A recorded example does not establish general model compatibility or a success rate.

Source reviewed · Upstream revision 8a1541c4a3ff

What this skill does

MCP servers expose tools an agent can invoke. This skill walks through schemas, error handling, and the boundary between a SKILL.md (instructions) and a server (side effects).

Prefer a skill when the knowledge is stable. Prefer MCP when the data changes or the action must actually run.

Intended uses

  • Stand up a typed MCP server for an internal API
  • Decide skill vs MCP vs plugin for a new workflow
  • Review tool schemas before shipping to a team

Implementation notes

  • Keep tool names specific; generic "run" tools are hard to choose.
  • Return errors the model can act on, not stack traces only.
  • Document required auth in the skill that wraps the server.

Try it, then tell us how it went

A copied prompt does not tell us whether the task worked. Report your result after checking the output. We collect your tool and outcome, never your files or prompt.