ProjectBuild a Permission-Aware Knowledge Hub as an MCP Server: the explanation · Module 16: Connecting Claude Code, Cursor and Claude Desktop · part 18 of 22
Part 18 · Module 16: Connecting Claude Code, Cursor and Claude Desktop
Module 16: Connecting Claude Code, Cursor and Claude Desktop
21 min read·
About 40 minutes · Needs: Module 3 (MCP transports: stdio and Streamable HTTP), Module 12 (OAuth sign-in and personal access tokens), Module 15 (the console's Connect page) · Code: console/src/pages/Connect.tsx, console/src/lib/mcpUrl.ts, server/wayfinder/mcp/stdio.py, source-code/README.md, research/claude-code-livetest.md
Where this fits
The parts this module covers: the three AI clients (amber, their models write the answers), the MCP Inspector, and the ways in (blue): nginx and /mcp over HTTP with OAuth, and the stdio bridge with a personal access token.
Everything so far was the server. This module is the other end: how an engineer's own AI client finds Wayfinder, signs in and starts calling its tools. Three clients are covered, Claude Code, Cursor and Claude Desktop, plus the MCP Inspector for debugging, and then the one real end-to-end run: five questions asked through Claude Code as Noor and Fatima.
The rest of this project is yours to keep
This project is bought on its own, once, and stays readable afterwards, including the parts added to it later.