Chrome Bridge

中文文档

Chrome Bridge is an open-source browser control bridge for Chrome and Edge. It lets AI agents operate the local browser: open pages, click, fill forms, extract content, take screenshots, export PDFs, and upload files.

Highlights:

• Local execution: page content, cookies, and login sessions stay on the machine.
• Browser reuse: works with the user's current Chrome/Edge profile and login state.
• Agent-friendly: Claude Code, Cursor, Codex, and any agent that can run HTTP or shell commands can use it.
• Fully open source: Go daemon, TypeScript extension, agent skill package, and website are auditable.

Installation

1. install
   macOS / Linux:

curl -fsSL https://github.com/zhaocore/chrome-bridge/raw/refs/heads/master/install.sh | bash

Windows PowerShell:

irm https://github.com/zhaocore/chrome-bridge/raw/refs/heads/master/install.ps1 | iex

2. install the browser extension:

Chrome Web Store - ChromeBridge

Verify the connection:

~/.chrome-bridge/bin/chrome-bridge status

It is ready when running: true and extension_connected: true are shown.

Use Cases

• E-commerce comparison: search across platforms and extract prices and links.
• Research: open multiple sources and collect structured information.
• Form filling: fill repeated forms automatically.
• Data entry: extract web data and enter it into systems.
• Job search: search and filter roles across job boards.
• Web archiving: batch screenshots or PDF exports.

Packages

Package	Stack	Role
chrome-bridge-api	Go 1.21 + Gin + WebSocket	Local daemon with HTTP/WebSocket APIs
chrome-bridge-plugin	TypeScript + Vite + Manifest V3	Chrome/Edge extension that runs browser actions
chrome-bridge-skill	Markdown	Agent skill package and usage guide
chrome-bridge-web	Next.js 16 + React 19 + Tailwind v4	Website

How It Works

AI Agent
  -> 127.0.0.1:10089 HTTP API
  -> local daemon
  -> ws://127.0.0.1:10089/ws
  -> Chrome/Edge extension
  -> Chrome DevTools Protocol
  -> current browser tab

Default listen address: 127.0.0.1:10089

The extension uses chrome.debugger to call Chrome DevTools Protocol. It inherits the current browser's login state, cookies, and extension configuration. Connect it only to a trusted local daemon.

Commands

~/.chrome-bridge/bin/chrome-bridge start
~/.chrome-bridge/bin/chrome-bridge stop
~/.chrome-bridge/bin/chrome-bridge restart
~/.chrome-bridge/bin/chrome-bridge status

List available tools:

curl -s http://127.0.0.1:10089/tools

Call example:

curl -s -X POST http://127.0.0.1:10089/command \
  -H 'Content-Type: application/json' \
  -d '{"action":"navigate","args":{"url":"https://example.com","newTab":true},"session":"demo"}'

Tools

Tool	Description
navigate	Open a URL
find_tab	Find and reuse an existing tab
snapshot	Read the page accessibility tree
click / mouse_click	Click an element
fill	Fill an input or editable area
evaluate	Run JavaScript
screenshot	Take a screenshot and return the file path
save_as_pdf	Export the current page as PDF
network	Read or analyze network requests
upload	Upload files
list_tabs	List session tabs
close_tab	Close the current tab
close_session	Close all tabs in a session
cdp	Call a CDP method directly
key_type / send_keys	Type text or send key events

Use GET /tools for the full parameter schema.

Sessions

Each session maps to a group of browser tabs. Use different sessions for different tasks to reduce interference.

Close a session after the task:

curl -s -X POST http://127.0.0.1:10089/command \
  -H 'Content-Type: application/json' \
  -d '{"action":"close_session","args":{},"session":"demo"}'

WebSocket Protocol

The daemon and extension communicate through WebSocket.

Message Type	Direction	Purpose
hello	extension -> daemon	Handshake
hello_ack	daemon -> extension	Handshake acknowledgement
ping / pong	Both	Heartbeat
tool_call	daemon -> extension	Tool call
tool_result	extension -> daemon	Tool result

tool_call example:

{
  "type": "tool_call",
  "requestId": "req-1",
  "payload": {
    "name": "navigate",
    "args": {
      "url": "https://example.com",
      "newTab": true
    }
  }
}

Directory Structure

chrome-bridge/
├── install.sh
├── install.ps1
├── AGENT_INSTALL.md
├── chrome-bridge-api/
│   ├── cmd/chrome-bridge/
│   └── internal/
├── chrome-bridge-plugin/
│   ├── src/background/
│   ├── src/popup/
│   └── public/manifest.json
├── chrome-bridge-skill/
│   ├── SKILL.md
│   └── references/operations.md
└── chrome-bridge-web/
    ├── app/
    ├── components/
    └── docs/

FAQ

Does Chrome Bridge upload page content?

No. Browser operations run locally. The agent's own model requests follow the rules of the selected model platform.

How is it different from Playwright or Selenium?

Playwright and Selenium are script automation frameworks. Chrome Bridge is built for AI agents and reuses the user's current browser environment.

Is it paid?

Chrome Bridge is free and open source. Cost depends only on the agent or model service you use.

Which agents are supported?

Any agent that can run HTTP requests or shell commands can connect. The installer tries to inject skill files for runtimes such as Claude Code, Cursor, and Codex.

References

• GitHub - chrome-bridge
• Chrome Web Store - ChromeBridge
• Chrome DevTools Protocol
• Chrome Extensions Manifest V3