Give Claude Code, Cursor or VS Code a real signed-in browser

How do you connect your editor to Minda?

Once per editor, in about a minute:

1. Open Minda on your computer, go to Settings, Developer and turn on Let coding agents control Minda. Keep Minda open while your editor uses it.
2. Under Connect an agent, pick your editor and press Copy. The command or config holds your computer’s own path to Minda’s MCP server, and the profile folder when yours is not the default one.
3. Run the command in a terminal (Claude Code, Gemini CLI, OpenClaw) or paste the config into your editor’s MCP settings (Codex, Cursor, VS Code, Antigravity, Windsurf, Claude Desktop). Node.js 18 or later must be installed.
4. Ask your editor’s agent, for example: In Minda, open my dashboard and read the error. Minda opens the page in a background tab with your sign-ins, and the tab you are on stays on screen.
5. While Ask before actions is on, as it is from the start, Minda asks you before each click, typed word or key press. Recent calls in Settings, Developer lists what each agent asked for, and New token cuts every agent off at once.

Which code editors connect to Minda?

Claude Code, Codex, Cursor, VS Code, Antigravity, Windsurf, Gemini CLI, Claude Desktop, OpenClaw and any other MCP client. Settings, Developer shows each command with your computer’s own path; below it reads <the path Settings, Developer shows>.

Claude Code

Run once in a terminal:

claude mcp add minda -- node "<the path Settings, Developer shows>"

Codex

Add to ~/.codex/config.toml (the Codex CLI, its IDE extension and the ChatGPT desktop app all read it):

[mcp_servers.minda]
command = "node"
args = ['<the path Settings, Developer shows>']

Cursor

Add to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "minda": {
      "command": "node",
      "args": [
        "<the path Settings, Developer shows>"
      ]
    }
  }
}

VS Code

In VS Code, run MCP: Open User Configuration from the Command Palette and add this; Copilot in agent mode then has the tools:

{
  "servers": {
    "minda": {
      "type": "stdio",
      "command": "node",
      "args": [
        "<the path Settings, Developer shows>"
      ]
    }
  }
}

Antigravity

In Antigravity, add to its MCP config, mcp_config.json (Settings → Customizations):

{
  "mcpServers": {
    "minda": {
      "command": "node",
      "args": [
        "<the path Settings, Developer shows>"
      ]
    }
  }
}

Windsurf

In Windsurf (now Devin Desktop), open the Cascade panel’s … menu → Open MCP config file, and add to mcp_config.json:

{
  "mcpServers": {
    "minda": {
      "command": "node",
      "args": [
        "<the path Settings, Developer shows>"
      ]
    }
  }
}

Gemini CLI

Run once in a terminal:

gemini mcp add -s user minda node "<the path Settings, Developer shows>"

Claude Desktop

In Claude Desktop, open Settings → Developer → Edit Config, and add to claude_desktop_config.json:

{
  "mcpServers": {
    "minda": {
      "command": "node",
      "args": [
        "<the path Settings, Developer shows>"
      ]
    }
  }
}

OpenClaw

Run once in a terminal while the OpenClaw gateway is stopped:

openclaw mcp add minda --command node --arg "<the path Settings, Developer shows>"

Other MCP client

Any other MCP client: add a local (stdio) server that runs node with this file. Most clients take this entry:

{
  "mcpServers": {
    "minda": {
      "command": "node",
      "args": [
        "<the path Settings, Developer shows>"
      ]
    }
  }
}

Command line

For scripts: "tools" lists the tools, "call <tool> '<json>' --task <name>" runs one.

node "<the path Settings, Developer shows>" tools

What can your editor do in Minda?

What a person does in a browser tab, with your sign-ins, in tabs of its own:

• Open pages and move between them: tabs, open, navigate, go_back, go_forward and close_tab.
• Read a page: read, snapshot (every link, button and field, numbered), extract (a table or list as data) and screenshot.
• Act on a page: click, type, press_key, scroll and fill_form, with your OK while Ask before actions is on.
• Search your history, bookmarks and saved pages, translate a page and search the web. Reading your own tabs and history always asks you first.
• Each task’s tabs sit together in a tab group named after the task. finish closes them, and a task left unused for 30 minutes closes them by itself.

What keeps it safe?

The bridge is off until you turn it on, and it only answers programs on your computer.

• Minda listens on 127.0.0.1 only, at a random port, and an agent needs the token in your Minda profile folder. A new token stops the old one at once.
• A request from a web page (one with an Origin header) is refused, so a site cannot drive Minda through it. Private windows are never reachable.
• Page text comes back to your editor fenced as untrusted data, with a note that it is not instructions: how Minda handles prompt injection.
• Recent calls keeps the tool, the site and the result, never page text or what was typed.
• A policy (Windows Group Policy, Intune, policies.json) or your team can keep it off for every member.

What are the limits?

• It is in Minda for desktop. Minda SI for Chrome has no MCP server: the bridge is the desktop app’s own.
• Minda must be running with the switch on, or the agent’s call fails and says so.
• Your editor starts Minda’s MCP server with Node.js, so Node.js 18 or later must be on the computer.
• Your editor’s model decides each step. Minda carries it out and asks you where you told it to.