Skip to main content

​
Set Up MCP with Cursor, Windsurf, or Any Other MCP Client

Claude Code gets fully automated setup (see the Claude Code guide). Every other MCP-compatible agent uses the same manual JSON config — Reikon generates the exact snippet for your machine, you paste it into the agent’s MCP settings.

​
1. Get your config snippet

In Reikon: Settings → MCP, pick your agent from the grid. Supported agents today are Cursor, Windsurf, GitHub Copilot, Gemini CLI, and Codex / AGENTS.md — or pick Other if your agent isn’t listed (the protocol is the same regardless). Reikon shows a ready-to-copy JSON block:
The command path is filled in automatically for your actual install — don’t type it by hand, it differs between a packaged app and running from source, and between macOS and Windows.

​
Headless (no desktop app)

You can host the same server from the rei CLI instead of the app — point command at rei with "args": ["mcp"]. See Run MCP Without the Desktop App for the full walkthrough.

​
2. Add it to your agent

Where this goes depends on the agent:
  • Cursor — if Reikon detects Cursor on your system, an Add to Cursor button appears in the panel. Click it and Reikon writes the config directly to ~/.cursor/mcp.json, merging with any existing entries. If you prefer to do it manually: Cursor → Settings → Features → MCP → Add new MCP server, paste the snippet.
  • Windsurf — similar location under MCP server settings; paste the snippet manually
  • GitHub Copilot — add the snippet to your Copilot MCP config
  • Gemini CLI — add the snippet to your Gemini CLI MCP config
  • Codex / AGENTS.md — add the snippet to Codex’s MCP config. The skill itself lands in the cross-agent AGENTS.md file, so any other agent that reads AGENTS.md picks it up too
  • Anything else — check that agent’s docs for “MCP server” or “stdio transport” configuration; the JSON shape above is the standard MCP config format, not Reikon-specific
Registering the server only tells the agent that Reikon’s tools exist. The skills are short blocks of instructions that help the agent actually reach for those tools at the right moment. Without them, agents often fall back to reading files and running git commands instead. There are two: the Reikon MCP guide (reikon — what each tool does and when to use it) and the Reikon queue worker (reikon-queue — turns “work Reikon’s dispatch queue” into the full pull → fix → mark_fix_returned → stop loop; it never resolves, so you keep the verdict). The queue worker is the MCP-pull companion to the desktop’s Copy prompt hand-off. For non-Claude agents, the skills write to the conventional rules file each agent already reads from your project root: In the panel for your agent, click Install skill. Reikon appends (or updates) a fenced Reikon block in that file, leaving the rest of the file alone — so it’s safe to run on a project that already has rules. When Reikon ships a new version of the skill, an Update available badge appears; click Install skill again to refresh just the Reikon block. Remove skill strips the block back out. Because these files live in your project root, you can commit them so your whole team gets the skill once the MCP server is registered locally.

​
4. Restart the agent if needed

Most MCP clients pick up new server config without a restart, but if tools don’t show up immediately, restart the agent.

​
Detection

Reikon’s header badge only shows a verified connection for Claude Code today — for other agents, there’s no automated “is it actually connected” check yet. Confirm it worked from the agent’s side: ask it to call get_project_health and see if it returns real data instead of saying the tool doesn’t exist.

​
Next

MCP tool reference for what’s available once connected.