Mcp Server Scaffolder

Everyone agrees agents need tools; almost nobody ships them cleanly. Most hand-rolled MCP servers start as a weekend script and calcify into an unauthenticated, undebuggable liability. This skill turns a plain-English description of the tools you want ("let my agent query our Postgres warehouse and post to Slack") into a production-ready MCP server scaffold — your choice of Python (FastMCP) or the TypeScript SDK — with authentication, structured errors, logging, and packaging done right from the first commit.

To take a natural-language description of desired agent tools and produce a complete, runnable MCP server project: tool definitions with JSON schemas, an auth layer, consistent error envelopes, request logging, a test harness, and deployment packaging (Dockerfile + README). The output is code a senior engineer would not be embarrassed to merge — not a demo snippet.

• When you need to give Claude Code, Cursor, or any MCP client access to an internal API, database, or service.

• When prototyping an agent integration and you want the scaffold to survive contact with production (auth, retries, observability) instead of being rewritten later.

• When standardizing how your team builds MCP servers — one scaffold shape, every time.

• When converting an existing ad-hoc script ("that Python file that hits the warehouse") into a proper MCP tool.

• When the "tool" is a single stateless function with no auth — a one-file script is fine; do not scaffold a server for it.

• When you need a hosted/remote MCP server with OAuth for third-party users — this skill scaffolds the server code; the OAuth client flow for connectors is covered by oauth-connector-blueprint.

• When wrapping a REST API you did not design — use rest-api-to-mcp-wrapper instead, which generates tools from an OpenAPI spec automatically.

• For MCP clients (building the agent side) — this skill builds the server side only.