Create MCP Servers

<objective> MCP servers extend Claude's capabilities by exposing tools, resources, and prompts. This skill guides creation of production-ready MCP servers with API integrations, OAuth authentication, response optimization, and proper installation in Claude Code and Claude Desktop. </objective>

<essential_principles>

<the_5_rules> Every MCP server must follow these:

1. Never Hardcode Secrets - Use ${VAR} expansion in configs, environment variables in code
2. Use cwd Property - Isolates dependencies (not --cwd in args)
3. Always Absolute Paths - which uv to find paths, never relative
4. One Server Per Directory - ~/Developer/mcp/{server-name}/
5. Use uv for Python - Better than pip, handles venvs automatically </the_5_rules>

<security_checklist>

• Never ask user to paste secrets into chat
• Always use environment variables for credentials
• Use ${VAR} expansion in configs
• Provide exact commands for user to run in terminal
• Verify environment variable existence without showing values
• Never hardcode API keys in code or configs </security_checklist>

<architecture_decision> Operation count determines architecture:

• 1-2 operations → Traditional pattern (flat tools)
• 3+ operations → On-demand discovery pattern (meta-tools)

Traditional: Each operation is a separate tool On-demand: 4 meta-tools (discover, get_schema, execute, continue) + operations.json </architecture_decision>

<context> MCP servers expose: - **Tools**: Functions Claude can call (API requests, file operations, calculations) - **Resources**: Data Claude can read (files, database records, API responses) - **Prompts**: Reusable prompt templates with arguments

Standard location: ~/Developer/mcp/{server-name}/ </context>

</essential_principles>

<routing> Based on user intent, route to appropriate workflow: