# Remote MCP Server Guide

> Build, deploy, and consume a remote MCP server (Python + FastMCP, Streamable HTTP). If you are an LLM agent
> helping a user wrap their *own* code as a remote MCP server, start with the agent playbook below — it has
> the build steps, the conventions to follow, and what to confirm with the user before writing code.

## For LLM agents

- [Agent playbook: build an MCP server from your code](https://badmintongpt-docs.nycu-adsl.cc/for-llm-agents/): step-by-step workflow, mandatory conventions (return JSON not exceptions, `Field`-described params, async job contract, file-as-URL, `/healthz`, auth), and the four decisions to confirm with the user (which functions→tools, sync vs async, auth model, hosting & domain).

## Docs

- [Guide](https://badmintongpt-docs.nycu-adsl.cc/): full reference — concepts, architecture, run locally, deploy remotely (Cloudflare Access **or** app-level Bearer token), connect a client, troubleshooting.
- [Example server](https://badmintongpt-docs.nycu-adsl.cc/example/): a complete, runnable reference implementation — sync tools, an async job (`start → status → result`) with progress, file delivery over HTTP, and a health check.

## Optional

- [Example server source (zip)](https://badmintongpt-docs.nycu-adsl.cc/example-mcp-server.zip): downloadable scaffold to copy and adapt.