Subconscious

Subconscious.ai / MCP

Your assistant.
Your next study.

Connect the AI assistant you already use to the Subconscious.ai Rehoboam MCP server. Work through a study design, review the draft, and inspect experiment results in your conversation.

Set up your clientGet your access token

Setup instructions for VS Code, Claude Code, Cursor, and other HTTP clients.

01 / Connect

One server.
Your account.

Sign in to Settings and select Get Access Token. Add a remote MCP server in your client using this address:

https://api.subconscious.ai/mcp/
Transport
Streamable HTTP
Authentication
Bearer access token

Keep your token in the client's private credential settings, never in a chat message or server URL. The setup guide has the exact configuration for each client.

Merge this into .vscode/mcp.json. VS Code prompts privately for your access token. Start the server and check that its tools appear.

{
  "inputs": [
    {
      "type": "promptString",
      "id": "subconscious-token",
      "description": "Subconscious.ai access token",
      "password": true
    }
  ],
  "servers": {
    "subconscious": {
      "type": "http",
      "url": "https://api.subconscious.ai/mcp/",
      "headers": {
        "Authorization": "Bearer ${input:subconscious-token}"
      }
    }
  }
}

Configuration checked against the public MCP setup guide on October 10, 2026. Review the study and cost before any paid launch.

Check that the server's tools appear before starting. Listing tools does not launch a paid experiment.

02 / Try it

Start with
a question.

Design a study

Help me design a job-offer preference study for US software engineers. Include salary, working arrangement, and commute. Use exact levels with units and keep benefits separate from salary. Ask about missing details and show me the draft. Do not launch until I review the design and approve the client's confirmation.

Read existing results

Help me inspect my completed job-offer study. Find it if study discovery is available; otherwise ask for its run ID. Verify the results, discover the exact attribute labels, and compare importance and two offer configurations I approve. Use direct numeric tools. Include sample sizes, units, uncertainty, and caveats. Do not launch a new experiment or call the delegated analyst.
Read the design and analysis guidance

03 / Review

You approve
the launch.

Review the experiment and its cost before approving a paid launch. Your client must support the server's confirmation flow. If it cannot show the approval request, use the browser app.

Follow the returned run ID after launch. If a read times out, retry the read instead of starting another experiment.

Read modeled preference shares as scenario comparisons, not observed market shares. Utility ratios are not dollar premiums. Keep sample sizes, uncertainty, and missing evidence with the result.

Explore the MCP tool reference