Ship an MCP Server in Python – Fast (Christoffer Noring)（Z-Library）

【One-Line Pitch】 A hands-on field guide for Python developers who want to build, test, and ship a real Model Context Protocol (MCP) server—from a minimal local prototype to a Streamable HTTP deployment with CI checks. Best for readers who already run Python and basic tooling and want a working server in editors and their own apps, not just theory. 【Book Arc】 - **Opening (~0%–13%)**: Establishes the mental model—host, client, server—and why MCP replaces bespoke glue with one standard surface for tools, resources, and prompts. - **Early (~13%–33%)**: Breaks down the three core primitives (Tools, Resources, Prompts), the initialize/initialized handshake, and transport choices (stdio for local, Streamable HTTP for remote, SSE deprecating). - **Middle (~33%–60%)**: Scaffolds a minimal Python server with UV, implements the flagship `add` and `joke` tools, and adds one static resource plus one prompt template. - **Late (~60%–73%)**: Runs the server locally over stdio and validates it visually in MCP Inspector, watching for a clean handshake. - **Ending (beyond ~73%)**: Excerpts do not cover the later chapters in detail—integration via mcp.json, a programmatic LLM client, Streamable HTTP migration with security, and the CI/rollout plan are listed in the table of contents but not expanded here. 【Key Takeaways】 - **MCP is "function calling on steroids"** (Opening): one clean, discoverable surface for tools, resources, and prompts that cuts integration cost and accidental complexity. - **Three primitives, three jobs** (Early): Tools are computation with crisp contracts; Resources are static, retrieved context; Prompts are reusable recipes. Sloppy schemas and vague descriptions cause tool misfires. - **The handshake is your mental anchor** (Early): `initialize` then `initialized` negotiates capabilities before normal calls. Skipping this model leads to hardcoded assumptions and breakage when servers evolve. - **Pick transport by stage** (Early): stdio for fast local iteration; Streamable HTTP for remote deployment; SSE still works but is being deprecated—plan a migration path. - **Descriptions matter more than function names** (Middle): Plain-language, verb-first tool descriptions are the bridge between how users speak and how the server operates. - **Keep the first loop tight** (Middle): Use UV to scaffold fast, keep `server.py` small, register tools/resources/prompts, and wire lifecycle hooks—nothing more. - **Test shape, not exact output** (Middle): For API-backed tools like `joke`, assert the call succeeds and returns non-empty text; don't assert the exact joke. - **Validate visually with MCP Inspector** (Late): Run over stdio via `mcp dev server.py` and watch for a clean handshake before blaming tool code. 【Reading Tips】 - **Deep-read Chapters 1–2**: The mental model and minimal server are the foundation; everything later builds on getting the handshake and primitives right. - **Skim the code listings, then type them**: The `add`/`joke`/resource/prompt examples are short—reproduce them in your own project rather than copy-pasting. - **Treat Inspector as your first debugging tool**: If tools don't list or calls silently no-op, check the handshake and transport wiring before touching tool logic. - **Plan for the later chapters**: Since excerpts don't cover mcp.json integration, the programmatic client, or Streamable HTTP security in depth, budget extra time there when you reach them. 【Coverage Limits】 This guide is based on stratified excerpts covering roughly the first three-quarters of the book; the later chapters on host integration, programmatic clients, Streamable HTTP migration, security, and CI rollout are named in the table of contents but not detailed in the source material.