Overview
BrowserPair connects AI assistants to the Chrome profile you already use. Instead of launching a fresh browser and asking you to sign in again, an agent can work with authorized websites in your existing browser environment, subject to the permissions and approvals you choose.
It supports AI apps such as ChatGPT and Claude through a browser connection built around the Model Context Protocol (MCP), alongside other integration options. Its job is not to make every browser action automatic. It is to make useful browser work possible without treating a signed-in account as an unlimited permission slip.
The problem
A new browser session is often ideal for tests. It is less useful when the task starts with a state that already exists: an authenticated dashboard, a saved search, a partially prepared form, or a workflow behind single sign-on.
Connecting an agent to a personal browser solves the context problem but creates a more important one: authority. Reading a page, editing a draft and submitting a payment are different operations. A system that can do all three needs to distinguish them.
Product approach
BrowserPair brings together a few ideas that matter more than a long list of automated clicks:
- Existing sessions: work from authorized pages in the user’s Chrome profile instead of forcing a second login.
- AI app connections: connect compatible assistants through MCP and other supported interfaces.
- User-controlled permissions: choose the browser and website access available to tasks, with configurable approval behavior for sensitive steps.
- Visible outcomes: review task progress, results and actions that may not have completed cleanly.
- Recovery over repetition: check an uncertain outcome before trying an action again.
For example, asking an agent to summarize the status of an authenticated dashboard is different from asking it to change the records on that dashboard. The same browser can support both; the authority should not be assumed to be the same.
What it is not
BrowserPair is not a replacement for Playwright or isolated testing browsers. For deterministic tests, repeatable CI jobs and disposable sessions, those tools are often a better fit.
It also does not mean browser data never leaves your machine. Signed-in sessions stay in Chrome, but information observed during an authorized task may be processed by the connected AI provider. Browser permissions, account settings and the sensitivity of the page still matter.
My role
I developed the product direction, browser-to-agent experience, permission and approval model, and engineering approach. The product grows from a question I keep coming back to: how do you give an AI assistant enough context to be genuinely useful without quietly giving it more power than the task requires?
Explore
The product website is BrowserPair. I also wrote about the broader design challenge in AI Agents in Your Logged-In Chrome: A Practical Guide.