agent-browser-mcp
335234131/agent-browser-mcp
MCP server that lets Hermes control your real, logged-in Chrome with CDP, screenshots and physical input
agent-browser-mcp is an MCP server that connects an agent to the Chrome you are already using, keeping logins, cookies and open tabs, and exposes page scanning, JavaScript execution, Chrome DevTools Protocol commands, screenshots and real mouse and keyboard input.
What agent-browser-mcp does
Rather than a sandboxed browser, agent-browser-mcp attaches to your own Chrome through an unpacked extension. A local bridge called TMWebDriver listens on 127.0.0.1 ports 18765 (WebSocket) and 18766 (HTTP), and the MCP service exposes tools such as list_tabs, switch_tab, open_url, scan_page, execute_js, cdp_command, cdp_batch, get_cookies, capture_page_screenshot, mouse_click, type_text and hotkey.
A command-line helper prints the extension path and a ready-made Hermes config snippet, and a doctor command reports extension, port and tab status as JSON. The README's Hermes config adds an agent_browser entry under mcp_servers in ~/.hermes/config.yaml, and examples are included for Claude Desktop and Cursor. It warns that mouse movement, clicks, typing and logged-in sessions are all real, so it should be used only with MCP clients and agent environments you trust. The README is in Chinese.
Key features
- Attaches to your real Chrome with its logins, cookies and open tabs
- Page scanning and in-page JavaScript execution
- Single and batch Chrome DevTools Protocol commands
- Page and desktop screenshots
- Real mouse, keyboard and hotkey input
- doctor command that reports setup status as JSON
When to use it
- Letting Hermes read a logged-in dashboard or feed without a fresh login
- Taking a CDP screenshot of a page the agent is working on
- Falling back to real clicks and typing when ordinary browser automation is blocked
Who it is for: Hermes Agent users who want their agent to work inside their everyday Chrome session, and who can read a Chinese-language README.
How it fits with Hermes Agent
The README is written around Hermes: it provides a print-hermes-config command, a mcp_servers entry for ~/.hermes/config.yaml and hermes mcp test agent_browser to verify it.
How to install agent-browser-mcp
These commands are copied from the project's README. Check the repository for the latest steps before you run them.
cd agent-browser-mcp
pip install -e .Requirements: macOS or Windows, Python 3.10+, Google Chrome with the unpacked extension loaded, and an MCP client such as Hermes Agent, Claude Desktop or Cursor
FAQ
What is agent-browser-mcp?
agent-browser-mcp is an MCP server that gives an agent control of your real Chrome browser. It supports page scanning, JavaScript, CDP commands, screenshots and physical mouse and keyboard input.
Does agent-browser-mcp work with Hermes Agent?
Yes, the README gives a Hermes config snippet for mcp_servers in ~/.hermes/config.yaml. You can check it with hermes mcp list and hermes mcp test agent_browser.
How do I install agent-browser-mcp?
Clone the repository and run pip install -e . inside it. Then load the unpacked Chrome extension from the path printed by agent-browser-mcp extension-path and open a normal web page.
Similar mcp & tools for Hermes Agent
All mcp & toolsWeb search MCP server that returns full page content for AI coding tools and agents like Hermes
asimons81 hermes-gptLocal-first MCP sidecar that exposes selected Hermes Agent capabilities to trusted MCP clients
sandbaseai SandBase CLILocal MCP bridge that connects 25 AI clients, including Hermes Agent, to SandBase models and APIs
agenticros AgenticROSROS 2 interface layer that lets AI agents, Hermes Agent included, control robots over MCP
YV17labs GhostDeskMCP server that gives AI agents a full desktop they can see, click and type on
aresyn Codex Control Plane MCPMCP server that turns Codex Desktop into a durable worker for long-running tasks
Related guides: What is Hermes Agent? · How to run Hermes Agent securely