Guide · Agent skill
How to use the Build MCP Server agent skill for Claude
Updated · written with AI from the project's own documentation and checked against our catalogue data
The build-mcp-server skill is a guide for building a Model Context Protocol (MCP) server, which is a small add-on that lets Claude use an outside tool or service. It is written for developers and teams who want to wrap an API or service so Claude can call it. After following this guide you will know how to install the skill, how to ask for it in plain language, and what the agent will ask you before it writes any code.
Build MCP Server Guideanthropics/claude-plugins-official · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Download the build-mcp-server skill ZIP from its page in this directory.
- 2Unzip into a skills folder. Unzip the file into ~/.claude/skills/ for personal use, or into .claude/skills/ inside your project.
- 3Or upload in the Claude apps. If you use the Claude apps, go to Settings, then Capabilities, then Skills, and upload the skill.
- 4Make your request. Ask in plain language, for example to build an MCP server that wraps a specific API.
- 5Answer the discovery questions. Reply to the agent's batched questions about the connected service, users, action count, input needs, and auth so it can recommend a deployment model.
What does the build-mcp-server skill make the agent do?
When the skill is active, the agent acts as a design guide rather than a code generator. Its first job is discovery. It asks about what the tool connects to, who will use it, how many distinct actions it needs, whether it needs mid-call user input or rich display, and what authentication the upstream service uses.
After those answers, the skill recommends one deployment style. The usual recommendation is a hosted remote server using streamable HTTP. For servers that must run on the user's machine, it points to a bundled local option (MCPB) or, for personal prototypes, a local stdio script. It also picks a tool-design pattern and points you to reference files or other specialized skills for the details.
- Asks about the connected service, intended users, action count, input or display needs, and auth type
- Recommends one deployment model: remote HTTP, MCPB, or local stdio
- Recommends a tool-design pattern: one tool per action for small APIs, or search plus execute for large ones
- Hands off to related skills such as build-mcp-app and build-mcpb when those fit
How do you install the build-mcp-server skill?
Download the ZIP file from the skill's page in this directory and unzip it. Place the resulting folder in one of two locations, depending on how you want to use it. A personal install goes in ~/.claude/skills/. A project-specific install goes in the .claude/skills/ folder inside your project.
If you use the Claude apps instead, you can upload the skill there. Open Settings, go to Capabilities, then Skills, and upload the skill from that screen.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
How do you trigger the skill with a natural request?
You do not need a special command. The skill is meant to activate when you ask for MCP work in ordinary language. Phrases that the skill description lists as triggers include asking to build an MCP server, create an MCP, make an MCP integration, wrap an API for Claude, expose tools to Claude, or make an MCP app. Discussing building something with the Model Context Protocol also counts.
Be specific about the service you want to connect. The skill asks questions anyway, so a clear first message lets it skip the questions you have already answered.
- "Build an MCP server that wraps our internal billing API."
- "I want to expose tools to Claude for our project tracker."
- "Help me make an MCP integration for a weather service and decide whether it should be hosted."
- "Make an MCP app that shows a searchable customer picker."
What should you have ready before you start?
The skill assumes some technical background. Familiarity with Node or Python is needed for any scaffolding it produces. The agent also needs to fetch the Claude connector documentation at https://claude.com/docs/llms-full.txt so its advice matches Claude's actual auth types and limits, so the agent must be able to reach that URL.
If you follow the Cloudflare Workers deployment path, you need a Cloudflare account. The skill's reference notes describe that path as the fastest route from nothing to a live https URL.
- Working knowledge of Node or Python
- Network access so the agent can fetch https://claude.com/docs/llms-full.txt
- A Cloudflare account, only if you use the Cloudflare Workers deploy path
What are the safety notes for scripts and network access?
The skill does not run scripts on your machine. It does use the network, because it fetches the Claude connector docs URL during its work. Expect the agent to reach that external address.
The reference files cover auth patterns such as OAuth with CIMD or DCR, and some auth types require contacting Anthropic's review address. Review the auth choice carefully before you publish a connector, and treat any credentials you handle as sensitive.
- No scripts are run by the skill itself
- Network use: yes, to fetch https://claude.com/docs/llms-full.txt
- Auth decisions are covered in references/auth.md; keep client secrets out of shared files
What tips help you get good results?
Answer the discovery questions in one message if you can. The skill is designed to batch its questions, and a complete answer lets it move to a recommendation sooner. If you already know you need a hosted server, say so, since the skill will still confirm the choice against your answers.
The skill is a guide, so it will not scaffold code until the design questions are answered. For a quick one-off script, the design discussion may add more time than value. The reference files hold the detailed patterns, so keep them in the skill folder next to SKILL.md.
- Give the service name, users, action count, and login method in your first message
- Keep all files in the unzipped folder, including the references directory
- Use the skill for hosted or shared servers more than for throwaway scripts
What people use it for
- Wrapping a cloud API such as a SaaS or REST service so Claude can call its operations as tools
- Deciding whether a team's MCP server should be hosted remotely or run locally on each machine
- Choosing between one tool per action and a search-plus-execute pattern for a large API
- Planning OAuth for an upstream service before writing server code
Frequently asked questions
How do I install the build-mcp-server skill?
Download the ZIP from the skill page and unzip it into ~/.claude/skills/ for personal use or into .claude/skills/ in a project. You can also upload it in the Claude apps under Settings, then Capabilities, then Skills.
Is the build-mcp-server skill free?
This page does not state a price for the skill. The skill is listed under the Apache-2.0 licence. A Cloudflare account is needed only if you follow the Cloudflare Workers deploy path.
What do I need to use the build-mcp-server skill?
You need familiarity with Node or Python for any code the skill produces, and network access so the agent can fetch https://claude.com/docs/llms-full.txt. A Cloudflare account is required only for the Cloudflare Workers deployment path.
Does the build-mcp-server skill run scripts on my computer?
No, the skill is listed as not running scripts. It does use the network, because it asks the agent to fetch the Claude connector documentation at a fixed URL.
Should I use this skill or an existing MCP server?
Use this skill only if you want to build an MCP server. If you just want to use an existing MCP server, this skill is not the right tool. It is also not suited to quick one-off scripts, since it asks design questions before scaffolding.
Why does the skill ask questions instead of writing code?
The skill treats discovery as the first job because the wrong server shape causes costly rewrites later. Its questions cover what the server connects to, who uses it, how many actions it needs, and the auth method, and the recommendation follows from those answers.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers and teams who want to wrap an API for Claude and need help deciding between a hosted remote server and a local one, plus guidance on tool design and OAuth before coding. It is a poor fit if you only want to use an existing MCP server, or if you need a quick throwaway script with no design discussion. Its reference files hold the detailed patterns, so keep the whole folder together.
Download the skill