TabBridge

Lets AI agents on your computer (Claude Code, Codex, Gemini CLI and others) use this browser, with your approval.

TabBridge lets an AI agent that runs on your own computer use your Chrome: Claude Code, Codex, Gemini CLI, Cursor, or any app that speaks the Model Context Protocol (MCP). TabBridge is only a bridge. It has no AI model and needs no API key. Your agent does the thinking; TabBridge gives it hands in the browser, in your normal profile with your sign-ins. What an agent can do • Open pages, go back and forward, reload • Read the page as a list of elements, find things, read the text • Click, type, press keys, fill in forms, scroll • Take screenshots, saved as a numbered step-by-step guide in your project • See what DevTools sees: console, network requests with headers and bodies, elements and styles, storage, cookies, IndexedDB, cache, service workers, performance and security • page_report: everything about a broken page in one call • Right-click any page or element and choose "Send to my agent" • Watch each step your agent takes in the live console (Chrome's side panel), and send it a note You stay in charge • Agents only use their own tabs, kept in a TabBridge tab group • YOLO mode is on at first; switch it off and risky clicks (send, buy, pay, delete, publish…) ask you first • Cookie values and tokens are hidden from agents unless you allow them • Optionally, be asked before an agent uses each new site • Everything stays on your computer: TabBridge has no server Setup 1. Add TabBridge to Chrome. 2. Run once in a terminal: npx @verbodhpteltd/tabbridge install 3. Connect your agent, for example: claude mcp add tabbridge -- npx -y @verbodhpteltd/tabbridge mcp Open source (Apache 2.0): https://github.com/verbodh-pte-ltd/tabbridge