Local browser bridge for agents

Let AI operate real Chrome.

Chrome Bridge turns agent instructions into browser-side actions. A local daemon, Chrome extension, and WebSocket bridge connect real tabs, so automation can use login sessions, screenshots, network visibility, and debuggable results.

chrome://bridge/session/demo

Current tool

snapshot → click → fill

$ navigate https://console.example.com

$ snapshot returns @e refs

$ fill @e42 campaign_id

$ screenshot ./review.png

Status

daemon

running: true

extension

connected

session

demo

Architecture map

Chrome Bridge architecture diagram

Capabilities

Do what actually happens inside the browser.

Real browser control

Navigation, clicks, typing, keys, and mouse events happen in the user's own Chrome tabs.

Page state reading

Accessibility snapshots return structured page trees and stable element references.

Screenshots and PDFs

Capture pages, elements, and PDFs to local files for review, archive, and delivery.

Network and CDP

Inspect requests and responses, then use Chrome DevTools Protocol when deeper debugging is needed.

File upload

Upload local files into web forms for admin setup, asset submission, and repeated workflows.

Session isolation

Run separate tasks in separate sessions and tab groups, keeping complex flows isolated.

Install

Install the daemon and browser extension.

Chrome Bridge has two local parts: a daemon and a Chrome extension. The daemon exposes the 127.0.0.1:10089 HTTP API, while the extension connects over WebSocket and executes actions in real tabs.

Local daemon

Receives agent commands, maintains sessions, forwards tool calls, and saves screenshot or PDF outputs.

Connection check

The status command reports both daemon health and extension connectivity, making installation issues easy to locate.

1

Step 1: Install the local daemon

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

Step 2: Install the browser extension

Chrome Web Store - ChromeBridge~/.chrome-bridge/bin/chrome-bridge status

When running: true and extension_connected: true both appear, agents can start calling browser tools.

Quick start

Three steps.

Confirm the local service, connect the extension, then let the agent send actions through the HTTP API. Status and logs keep failures diagnosable.

1

Start the local daemon

The daemon listens on 127.0.0.1:10089 and receives agent commands.

~/.chrome-bridge/bin/chrome-bridge start
2

Connect the Chrome extension

The extension connects over WebSocket and performs actions inside the browser.

~/.chrome-bridge/bin/chrome-bridge status
3

Call tools from the agent

The agent sends actions like navigate, snapshot, click, fill, and screenshot.

POST http://127.0.0.1:10089/command

Use cases

Built for high-friction web workflows.

This is not a headless-browser replacement. It is for agent automation that needs real login state, real pages, and local control boundaries.

Admin workflow setup

Open real admin consoles, read existing settings, upload assets, fill campaign fields, and capture review screenshots.

Less repetitive clicking with auditable output.

Authenticated web automation

Reuse the user's current login state for pages behind SSO, internal permissions, or post-login checkpoints.

No need to hand credentials to scripts.

Page review and research

Read page structure, inspect network traffic, save PDFs, and turn complex webpages into structured results.

Useful for QA, archival, and competitive research.

E-commerce price and stock checks

Search products across sites, read prices, promotions, inventory, and detail links, then turn them into comparable results.

Reuse real accounts and localized pages.

Job search filtering

Open job boards, filter roles by keyword, location, compensation, and company details, then keep source links.

Turn repeated searches into structured lists.

Web archive and delivery

Capture screenshots or export PDFs in batches so key pages are preserved at a specific point in time.

Deliverables land directly as local files.

Architecture

Clear boundaries. Short path.

The agent calls a local HTTP API. The daemon validates tool arguments, maintains session state, and forwards work to the extension. The extension executes inside Chrome and sends results back.

Agent
HTTP API
WebSocket Bridge
Chrome Extension
Real Browser Tab

Page actions

navigate / click / fill / send_keys / mouse_click

Page reading

snapshot / evaluate / network / cdp

File output

screenshot / save_as_pdf / upload

Tab management

find_tab / list_tabs / close_tab / close_session