Guide
How to let Claude Code or Codex use your real browser
Coding agents usually get a fresh headless browser with no logins, so anything behind a sign in is out of reach. Tether connects them to the browser you already use instead, one tab at a time, with your permission. This guide covers why that matters, how to set it up, and what to ask your agent once it's connected.
Why your own browser
Most of the work people want help with lives behind a login: a company ERP, a billing dashboard, a CMS, a university portal, an inbox. A headless browser started by Playwright or Puppeteer has none of those sessions, and handing an agent your password is a bad idea. Your everyday browser already has the cookies, the saved sessions and the extensions. Tether lets an agent use that browser without ever seeing a password.
How it works
Tether has two parts. A small local host runs on your machine and speaks the Model Context Protocol (MCP), so any MCP client can call it. A browser extension for Chrome, Brave, Edge, Firefox or Zen connects to that host and carries out what the agent asks, but only in tabs you shared or approved.
The agent reads each page as an outline of headings, text and controls with short references, then clicks, types or chooses by reference. Every action returns the updated page, so it always knows what happened. In Chromium browsers the clicks and keys arrive as real input.
Setup in four steps
- Install the host. Run
npm install -g tether-browser, thentether setup. It prints a pairing key and the exact lines for your agents. - Add Tether to your agent. Paste the line
tether setupprinted. For Claude Code it starts withclaude mcp add --scope user tether. Codex, Cursor and Gemini CLI take the same command as a stdio MCP server. - Install the extension. Get it from the latest release. In Firefox or Zen, open the signed
.xpi; no key needed. In Chrome, Brave or Edge, unzip it, load it as an unpacked extension and paste the key. - Check it. Run
tether doctor. It checks each piece and tells you what to fix.
Things to ask your agent
- “Open my invoices dashboard and export this month's unpaid ones as a CSV.”
- “Fill this job application with the details from my resume.”
- “The checkout on staging throws an error. Reproduce it and read the network log.”
- “Make this admin panel dark and easier on the eyes, like the screenshot I sent.”
- “Tell me when the price of this item drops below 2,000.”
- “Write a userscript that fills my timesheet with 9:30 to 18:00 every Friday.”
Staying in control
Nothing is visible to an agent until you click Share this tab or approve its request for a window of its own. A gold cursor with the agent's name shows every click before it happens, and Pause holds every action until you resume. When the agent reaches a CAPTCHA, a one-time code or a payment confirmation, it hands the tab to you and waits. The host listens only on your machine and there's no Tether server. Your agent sends what it reads to its own model provider, as it does with your code.
Tether or a headless browser
Use a headless browser for tests that run in CI. Use Tether when the job needs your real sessions or you want to watch the agent work.
Have an idea for Tether?
Tell us what you want your agent to do in your browser. The roadmap comes from the community.