mcp_skill

Overview

This skill implements MCP (Model Context Protocol) to build AI-native servers that expose tools, resources, and prompt templates for LLM integration. It provides TypeScript and Python SDK examples and ready-to-run patterns for STDIO, SSE, and HTTP transports. Use it to connect Claude Desktop or other clients to file systems, APIs, databases, and custom tools with minimal boilerplate.

How this skill works

The skill defines a Server that lists available tools, resources, and prompts and handles requests for tool execution, resource reads, and prompt retrieval. Tools are callable functions (e.g., read_file, list_directory), resources provide read-only data via URIs, and prompts return templated messages. Transport adapters (stdio, SSE, HTTP) expose the server to clients like Claude Desktop and support async operations and error handling.

When to use it

• Expose local files, project docs, or config to an LLM securely via MCP.
• Add executable helper functions (tools) for code review, file reads, or API queries.
• Provide reusable prompt templates for repeated LLM workflows (code review, summarization).
• Integrate an LLM client (Claude Desktop) with custom tooling over STDIO or HTTP.
• Rapidly prototype AI-native server capabilities in TypeScript or Python.

Best practices

• Define clear input schemas for tools so the LLM can validate arguments before calling.
• Return structured TextContent and set isError flags for graceful LLM error handling.
• Use async I/O (axios, aiosqlite, async fs) for nonblocking external calls and queries.
• Expose only necessary resources and sanitize file paths when serving file:// URIs.
• Load secrets and config from environment variables and avoid hardcoding keys.

Example use cases

• Filesystem server that lists directories and returns file contents for code inspection.
• API proxy tool that fetches remote data and presents formatted JSON to the LLM.
• Database query tool that runs read-only queries and returns results as JSON.
• Prompt library that supplies code_review templates for automated review workflows.
• Claude Desktop integration to surface project resources and interactive tools in the client menu.