All docs
MCP

Connect an AI client to Radinode over MCP

Give Claude, Cursor, or another AI client read-only access to your Radinode workflows, plans, and issues over the Model Context Protocol (MCP).

Radinode can act as an MCP server. Once connected, an AI client — Claude Code, Claude Desktop, Cursor, and others — can answer questions about your own workspace in plain language: "list my workflows", "what systems are in the Riverside integration", "which risks are still open on that plan". This guide walks through connecting a client.

What you get (and what you don't)

  • Read-only. Clients can read your workflows, integration plans, issues, risks, clients, devices, and partners. They cannot create, edit, or delete anything.
  • Scoped to your workspace. A key only ever returns data from the organization it was created in.
  • No PHI or secrets. Contact details, questionnaire answers, and share tokens are never exposed over MCP.
  • Plan. MCP access is a Pro feature.

Step 1 — Create an API key

In Radinode, go to My Account → MCP, give a key a name (e.g. "My laptop"), and select Create key. The secret is shown once — copy it now; you can always create another if you lose it. The same dialog shows a ready-to-paste configuration for each client, with your key already filled in. When in doubt, copy the block for your client straight from there rather than hand-editing the examples below.

Step 2 — Connect your client

Claude Code (CLI)

One command registers the server with your key:

claude mcp add --transport http radinode <your-server-url>/mcp \
  --header "Authorization: Bearer <your-key>"

Then ask Claude to "list my Radinode workflows".

Cursor / Windsurf

These clients support a remote MCP server directly. Add this to the client's MCP configuration:

{
  "mcpServers": {
    "radinode": {
      "url": "<your-server-url>/mcp",
      "headers": { "Authorization": "Bearer <your-key>" }
    }
  }
}

Claude Desktop

Claude Desktop loads local (stdio) servers only, so it connects through the mcp-remote bridge. This needs Node.js installed (check with node -v). Open Settings → Developer → Edit Config and add radinode as a top-level mcpServers entry — not nested inside any other object:

{
  "mcpServers": {
    "radinode": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "<your-server-url>/mcp",
        "--header", "Authorization:${AUTH_HEADER}"
      ],
      "env": { "AUTH_HEADER": "Bearer <your-key>" }
    }
  }
}

Passing the header through the AUTH_HEADER environment variable (note there is no space after the colon in Authorization:) avoids a known parsing bug with the space inside Bearer <key>. Fully quit Claude Desktop (⌘Q) and reopen it, then check Settings → Developerradinode should show as running.

Browser clients (claude.ai)

The web version of Claude connects through a custom connector over OAuth — no API key required. In Claude, open Settings → Connectors → Add custom connector, paste your <your-server-url>/mcp URL, and connect. You'll be sent to Radinode to sign in and approve read-only access; after that Claude reconnects automatically. The connection is scoped to your workspace and can be removed at any time from the connector settings.

What you can ask

Once connected, the client can call these read tools:

Ask for…What it returns Your workflowsNames, status, last updated A specific workflowIts systems and the connections between them Your integration plansNames, status, version, key dates A plan's statusPhase progress, open issues, and open risks Clients, devices, partnersHigh-level lists (no contact details)

Managing keys

Each key is personal to the person who created it and appears in My Account → MCP with its prefix and last-used date. Revoke a key there at any time — access stops on the client's next request. If a key is ever exposed, revoke it and create a new one.