Connect

Connect Rafter to Codex

Codex speaks stdio, Rafter is a hosted server — mcp-remote bridges the two. One config block and Codex can search your team's shared memory and save learnings back.

You need a Rafter account — create one if you haven’t.Takes about 3 minutes.No API keys — sign-in is OAuth, in the browser.
  1. Add Rafter to config.toml

    Add this to ~/.codex/config.toml.

    [mcp_servers.rafter]
    command = "npx"
    args = ["-y", "mcp-remote", "https://app.heyrafter.xyz/api/mcp"]

    npx needs Node 18+ on your PATH.

  2. Restart Codex and sign in

    Restart codex. On the first Rafter call the sign-in link is printed — open it, authenticate, and pick the workspace to connect.

  3. Ask something your team already knows

    Try “Search Rafter for our release checklist and follow it.”

Your first five minutes

The connection is the easy part. The habit is what pays: make Codex check the shared brain before it improvises. Three things to try in Codex:

To make it automatic, paste this line into Codex’s project or system instructions:

Check Rafter first: search our skills and memories before you answer, and save anything durable back with add_memory.

A longer startup prompt and a ready-made Claude Code skill live in the rafter-mcp repo.

Troubleshooting

Why the mcp-remote bridge?

Codex talks to MCP servers over stdio; Rafter is a hosted HTTP server. mcp-remote is the standard bridge between the two and handles the OAuth flow.

Which workspace should I pick when signing in?

The one whose skills and memories you want this tool to see. A connection sees exactly what you can see — your teams and your role — never more. You can connect again later and grant a different workspace.

Is my team's content used to train AI models?

No. Rafter does not use your content to train generalized or foundation models — see the privacy policy.

Give your whole team the same brain.From $9.99 per user — see pricing.

Create your account