LOCAL BROWSER CONTROL FOR MCP CLIENTS

Use the Chrome session
you already trust.

Browser Agent Bridge lets local AI agents work with your existing Chrome tabs and authenticated sessions, while keeping the control path on your machine.

ONE LOCAL CONTROL PATH

MCP clientstdio MCP serverloopback bridgeChrome extensionChrome

No cloud relay. No cookie export. No clean automation profile to sign into again.

Existing sessions

Operate tabs that are already open in your Chrome profile, including SSO-backed and internal applications available in that profile.

Agent-friendly targeting

Get an accessibility-oriented snapshot, then interact through stable IDs instead of brittle coordinates or website-specific scripts.

Modern web UI support

Target open shadow roots and same-origin frames, set framework-controlled inputs, and use trusted CDP input for rich editors.

Conservative defaults

Loopback-only binding, a required strong token, and disabled page evaluation help keep the integration explicit and local.

A DIFFERENT JOB THAN TEST AUTOMATION

Not another Playwright replacement.

Playwright and Puppeteer excel at repeatable tests in controlled browser profiles. Browser Agent Bridge is for developer agents that need a consented path into the user’s active browser context.

THREE STEPS

  1. Start a local bridgeBRIDGE_TOKEN=... npm start
  2. Pair the Chrome extensionLoad unpacked, then enter the local host, port, and token.
  3. Register the stdio MCP serverAdd a standard MCP configuration to your preferred client.
Read the complete setup guide →

Powerful by design

A connected agent can read and act in the browser context you expose. Only connect trusted MCP clients, review their actions, and never paste secrets into prompts or tool arguments.

Read the security model →