Firefox Browser MCP — Server

An MCP server that lets an LLM client drive a real Firefox browser — every tab, its DOM, arbitrary CSS selectors, and its network/API traffic. It talks to the companion firefox-extension/ WebExtension over a local WebSocket bridge.

MCP client (Claude Desktop, etc.)
        │  stdio (MCP)
        ▼
firefox-browser-mcp  ──ws://127.0.0.1:9010──►  Firefox extension  ──►  All tabs

Run with uvx

uvx --from ./mcp-server firefox-browser-mcp

Once published to PyPI: uvx firefox-browser-mcp.

CLI options

Connection settings can be passed as arguments (they override the FBMCP_* env vars):

uvx firefox-browser-mcp --host 127.0.0.1 --port 9010 --log-level INFO
uvx firefox-browser-mcp --version
uvx firefox-browser-mcp --help

The Firefox extension connects to this bridge; use the extension popup's Enable/Disable toggle to turn the connection on or off.

Configure your MCP client

{
  "mcpServers": {
    "firefox-browser": {
      "command": "uvx",
      "args": ["--from", "/absolute/path/to/mcp-server", "firefox-browser-mcp"]
    }
  }
}

Targeting tabs

Every tab has a stable numeric id (plus title and url). Get them with browser_list_tabs. Most tools accept an optional tab argument:

Environment variables

If you change the port, update the bridge URL in the extension popup.

Tools

Tabs — browser_list_tabs (id/title/url of every tab, all windows), browser_select_tab, browser_new_tab, browser_close_tab.

Navigation — browser_navigate, browser_go_back, browser_go_forward, browser_reload.

Inspection & extraction — browser_snapshot (accessibility tree with ref ids), browser_query (any CSS selector → refs, text, attributes, html), browser_get_text, browser_get_html, browser_get_attribute, browser_eval (run JS and return the result), browser_get_url, browser_screenshot, browser_get_console_logs.

Network / API — browser_get_network (captured requests with method, url, status, content-type, request & response bodies), browser_clear_network.

Interaction — browser_click, browser_type, browser_hover, browser_select_option, browser_press_key, browser_scroll, browser_wait. Interaction tools accept either a ref (from snapshot/query) or a CSS selector.

Status — browser_status.

Examples (what an agent can do)

• "List my tabs, then read the article in the tab with 'wikipedia' in its url": browser_list_tabs → browser_get_text(tab="wikipedia").
• "Grab all product prices": browser_query(selector=".price") or browser_eval(script="return Array.from(document.querySelectorAll('.price')).map(e=>e.textContent)").
• "Show the API calls this page made": browser_get_network(filter="xmlhttprequest") (includes JSON response bodies).
• "Log into the form": browser_type(selector="#email", text=...), browser_type(selector="#password", text=..., submit=True).

Development

cd mcp-server
uv venv && source .venv/bin/activate
uv pip install -e .
python -m firefox_browser_mcp.server