Siigo MCP Server

--- name: technical description: MCP/FastMCP best practices, progressive disclosure, middleware, context management. Use when working on MCP architecture, tool design, or Logfire observability. --- # Technical MCP Expert Research-First Implementation skill for MCP/FastMCP best practices. ## When to Use Invoke this skill when working on: - MCP server architecture decisions - Tool design and progressive disclosure - Context management optimization - FastMCP middleware and composition - Logfire observability patterns ## Key Resources ### FastMCP - [FastMCP Docs](https://gofastmcp.com/) - Official documentation - [FastMCP GitHub](https://github.com/jlowin/fastmcp) - Source and examples - [Tool Transformation](https://www.jlowin.dev/blog/fastmcp-2-8-tool-transformation) - Making tools LLM-friendly - [Middleware](https://www.jlowin.dev/blog/fastmcp-2-9-middleware) - Cross-cutting concerns ### MCP Protocol - [MCP Best Practices](https://modelcontextprotocol.info/docs/best-practices/) - Architectural guidelines - [MCP Spec](https://modelcontextprotocol.io/specification/2025-11-25) - Latest protocol specification ### Progressive Disclosure - [Lazy MCP Proxy](https://github.com/voicetreelab/lazy-mcp) - 85-93% context reduction - Pattern: Discover → Load → Execute (don't expose all tools upfront) - Use gateway pattern for 5+ related capabilities ## FastMCP Patterns ### Server Composition ```python # Dynamic composition (live links) mcp.mount("customers", customer_server) # Static composition (snapshot) mcp.import_server(shared_utils) ``` ### Dependency Injection ```python @asynccontextmanager async def get_client() -> AsyncIterator[Client]: client = Client() try: yield client finally: await client.close() @mcp.tool async def my_tool(client: Client = Depends(get_client)) -> str: return await client.fetch() ``` ### Context Usage ```python @mcp.tool async def long_task(query: str, ctx: Context) -> str: ctx.log.info(f"Processing: {query}") ctx.report_progress(0.5) return "result" ``` ## Logfire Integration ```python import logfire logfire.configure() # Automatic instrumentation for: # - HTTP requests (httpx) # - Async operations # - Custom spans with @logfire.instrument ``` ## Prompt When invoked, research the specific technical question using the resources above, then provide: 1. Current best practice recommendation 2. Code example if applicable 3. Links to relevant documentation