JovaniPink/mcp-browser-use

git clone https://github.com/JovaniPink/mcp-browser-use.git && cd mcp-browser-use
uv sync
uv run mcp-browser-use

Usage

• •Requires Python 3.11+, Google Chrome or Chromium, and uv.
• •Copy sample.env.env to .env and set the LLM provider you plan to use (e.g. MCP_MODEL_PROVIDER=anthropic with ANTHROPIC_API_KEY).
• •Register the server in Claude Desktop via mcpServers with command uvx and args [mcp-browser-use].
• •Debug with the MCP Inspector: npx @modelcontextprotocol/inspector uv --directory /path/to/project run mcp-browser-use.

Key features

• ✓Single run_browser_agent MCP tool over FastMCP that orchestrates the whole browser session and returns the final result.
• ✓Centralized create_browser_session builds BrowserSession from env vars — persistent profiles, proxies, and custom Chromium flags.
• ✓CustomAgent wraps browser-use with history export, richer prompts, and consistent error handling.
• ✓Input trimming/bounds checks, constrained step limits, proxy settings omitted from debug logs, and CDP credential redaction.

Best for

Clients (Claude Desktop and others) that need broad browser automation — form filling, clicking, screenshots, tab control — driven by natural language.

Caveats

• ⚠Beta software.
• ⚠Controlling a full browser instance remotely can grant broad access to the host machine — review documentation/SECURITY.md before exposing to untrusted environments.
• ⚠The Python 3.14/browser-use dependency migration is held pending upstream patching of transitive versions.
• ⚠Requires at least one LLM provider API key and a local Chrome/Chromium.