What it does

Attaches to a real Chrome profile instead of launching a headless one, so the agent sees the pages you are logged into. That is the whole value and the whole risk: it acts as you, in your session. Fine for reviewing a staging site; wrong for anything you would not click yourself.

Install

Terminal

npx -y @browsermcp/mcp

Client configuration

Paste this into your MCP client config (Claude Desktop, Cursor, or any client that reads mcpServers), then restart the client.

mcp.json

{
  "mcpServers": {
    "browser-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@browsermcp/mcp"
      ]
    }
  }
}

Use it for

Driving the browser you already have open, with your sessions and logins

Skip it if

Unattended or parallel runs; it controls one real browser

When to use it

Reach for this when the task needs browser inside the agent loop. Skip it if a plain shell command or a one-off script is enough — every server you add costs context and adds a failure mode.

Other TypeScript servers

Filesystem — Gives an agent scoped read, write, and search access to directories you explicitly allow
Memory — A persistent knowledge graph stored in a local JSON file, exposed through entities, relations, and observations
Sequential Thinking — Gives the model a scratchpad for structured, revisable reasoning: numbered thoughts that can branch, revise, and conclude