AI & MCP

MCP Setup

MCP (Model Context Protocol) lets external AI assistants control your live Recense session. The assistant can inspect datasets, build tables, and work with the canvas using the same tools as the in-app agent.

AI & MCP

How it works

Your MCP client connects to Recense, which routes tool calls to the browser session you have open. The client works with the same session you see — not a separate copy.

Browser tab must stay open

The browser tab must stay open. Closing it disconnects the MCP client.

AI & MCP

Setup (all clients)

The setup flow is the same for all MCP clients:

1

Open Recense in the browser, sign in, and load a dataset.

The MCP client needs a live session to connect to.

2

Generate an API key in Account → API Access.

The relay uses your Recense API key for authentication.

3

Add the Recense MCP server in your client using the relay endpoint.

Use the endpoint URL below and send your API key as a Bearer token.

4

Test with a read-only request.

Try "describe the dataset" or "list variables" to confirm the connection works.

Relay MCP endpoint 
https://mcp.recense.ai/mcp

AI & MCP

Per-client notes

Claude Code — add a new MCP server pointing to the endpoint above. Claude Code handles connection setup. Ideal for users who want Recense tools alongside their coding workflow.

Claude Desktop — add the endpoint in Claude Desktop's MCP server settings. Authentication is discovered automatically — no extra configuration needed.

ChatGPT Desktop — same endpoint as above. Start with read-only prompts (dataset description, variable listing) before moving to analysis actions.

AI & MCP

Custom clients

Supported today:

  • Live session relay — your client controls a signed-in browser session with full app access.
  • Headless CLI — direct access against a .sav file without a browser. Analysis engine only (no workspace UI).
Headless pattern 
recense-cli mcp --file /path/to/data.sav --name "My Survey"

Hosted compute

Enterprise hosted compute is available for organizations that have been enabled for it and have explicitly published datasets. It is separate from the live-session browser relay path.

AI & MCP

Available tools

All MCP clients can:

  • Inspect dataset structure, variables, and metadata.
  • Build and modify tables on the canvas.
  • Apply filters, weights, and measures.
  • Read analysis results.

Relay vs headless

The relay option gives full live-session app access. The headless CLI gives direct analysis engine access without the workspace UI.

AI & MCP

Troubleshooting

  • Client can't connect: verify the endpoint URL ends in /mcp, not just the site root.
  • Tool calls fail: check the Recense browser tab is still open and signed in.
  • Sign-in keeps looping: confirm you're signing in with the same account you expect the client to control.
  • Tier issue: verify your Recense plan includes MCP access.