Connect MCP

Plug your AI client into Brain so it can read and write your workspace.

What is MCP?

MCP (Model Context Protocol) is a way for AI clients to talk to external tools and data. Brain ships an MCP server that gives your client safe, scoped access to the same things you can already do in the UI: list databases, create rows, edit docs, run agents. Once your client is connected, you can ask it to “make a database for tracking deals” and it actually does it.

Get your token

prereq

1
Open Settings, then Developer in Brain.
2
Click "Create new token" and give it a name (e.g. "Claude Desktop").
3
Copy the token. You'll only see it once.

Treat the token like a password. Anyone with it can read or write everything in your workspace.

Configure your client

Pick your client below and paste the snippet into its config. Replace YOUR_BRAIN_TOKEN with the token you just copied.

Add to ~/.claude.json (or run "claude mcp add" and paste the URL plus header).

{
  "mcpServers": {
    "brain": {
      "url": "https://team-brain.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_BRAIN_TOKEN"
      }
    }
  }
}

Claude Code speaks MCP over HTTP directly, so it can hit the Brain endpoint with a single Bearer header. After saving, restart the Claude Code session for the new server to show up in /mcp.

Test it works

After restarting your client, paste this prompt to confirm the connection.

Use the Brain MCP to list my databases.

Your client will call s16_list_databases and read back the names of every database in your workspace. If you see them, you're connected.

Pair with the Brain skill

A single SKILL.md file your AI client reads on connect. Once it knows the s16_ API, the conventions, and the common workflows, you can stop prompting it on every turn. Drop the file into Claude Code, Claude Desktop, or Cursor and let your assistant move things around with confidence.

Full API reference

Every s16_ tool exposed over MCP, with arguments and return shapes.

Agents

Run the same tools automatically from a sandboxed script on a trigger.

Last updated June 24, 2026