Docs

MCP for agents

Stroke ships a built-in MCP (Model Context Protocol) server. Point an agent at it and it can read your schema and run queries against a database you already trust, with the connection and credentials staying on your machine.

01

Turn on the MCP server

Open Stroke → MCP, pick the connection you want to expose, and toggle the server on. Stroke serves MCP locally and shows the exact endpoint plus a copy-ready config for each client. Nothing is exposed to the network beyond your machine.

02

Connect Claude Desktop

Add Stroke to Claude's MCP config, then restart Claude. Use the snippet Stroke gives you (the port may differ); it looks like this:

claude_desktop_config.json
{
  "mcpServers": {
    "stroke": {
      "url": "http://127.0.0.1:4319/mcp"
    }
  }
}

The config file lives at:

macOS
~/Library/Application Support/Claude/claude_desktop_config.json
Windows
%APPDATA%\Claude\claude_desktop_config.json

After restarting, Claude lists stroke under its tools and can start querying.

03

Connect Cursor and others

In Cursor, open Settings → MCP → Add server and paste the same endpoint, or drop it into a project's .cursor/mcp.json. Any MCP-compatible client works the same way: point it at the URL Stroke shows and reload.

04

What agents can do

Once connected, an agent can inspect tables, columns, indexes, and relationships, then run read queries to answer questions or draft SQL for you. It works against the live schema, so answers reflect your actual database, not a guess.

05

Staying in control

The MCP server only runs while Stroke is open and you've enabled it for a connection. Turn it off any time from the same panel. Because everything runs locally, your credentials and query results never leave your device. See the Privacy Policy for the full picture.