Build MCP Server

## Setup: Track this exercise with Beads

```bash
bd create "Praxis: Build MCP server for [domain]" --type=feature --labels=learn,advanced
bd create "Design tool composition—what operations merge?" --type=task --labels=learn
bd create "Define JSON schema for tool inputs" --type=task --labels=learn
bd create "Implement stdio transport handler" --type=task --labels=learn
bd create "Implement tool execution logic" --type=task --labels=learn
bd create "Test with Claude Code integration" --type=task --labels=learn
```

---

Help me build MY own MCP server for [DOMAIN].

I'm creating a custom MCP server that extends Claude Code for my specific use case. Guide me through:

1. **Identify composed operations**:
   What operations do I frequently chain together?
   - Read config → validate → apply?
   - Query database → transform → cache?
   - Fetch API → parse → update local state?

   List 2-3 operations that should become single tools.

2. **Design tool schemas**:
   For each tool, define:
   ```typescript
   {
     name: "my_tool",
     description: "What this tool does",
     inputSchema: {
       type: "object",
       properties: {
         // Define inputs
       },
       required: ["..."]
     }
   }
   ```

3. **Set up the project**:
   ```bash
   mkdir my-mcp-server && cd my-mcp-server
   npm init -y
   npm install @modelcontextprotocol/sdk
   ```

4. **Implement the server**:
   Create src/index.ts with:
   - Server initialization
   - Tool registration
   - stdio transport handler
   - Tool execution logic

5. **Handle the protocol**:
   - Parse JSON-RPC requests
   - Return properly formatted responses
   - Handle errors gracefully

6. **Test locally**:
   - Add to Claude Code settings
   - Invoke tools and verify output
   - Debug any issues

7. **Document MY MCP server**:
   - What problem does it solve?
   - How do I use each tool?
   - What are the gotchas?

Domain: [YOUR_DOMAIN - e.g., "project management", "data pipeline", "content publishing"]
Operations to compose: [YOUR_OPERATIONS]