PagePilot is an MCP browser agent. A Chrome extension is paired to your account, and any MCP client connects through one connector URL; AI clients such as Claude Code and Claude Desktop can then operate in the tabs you allow. It was created by Sakthi Vignesh and is live at page-pilot-app.vercel.app.
Why would an AI agent need to use a browser?
Because plenty of useful software has no API. A supplier portal, a government form, a legacy admin screen or a SaaS tool on a plan without API access can only be used by a person clicking around. If an agent is going to help with that work, it has to be able to click around too. The alternative, giving the agent your username and password, is worse in almost every way.
What can PagePilot do?
- Work with tabs and navigation
- Read pages, click, type and fill forms
- Upload files and take screenshots
- Inspect browser storage
All of this is exposed over MCP, so the same toolset is available to any client that speaks the protocol. New to the term? Start with what an MCP server is.
How do you keep a browser agent under control?
This is the part we care about more than the toolset. PagePilot ships with four controls:
- Approval gates for risky actions, so the agent pauses and asks.
- Site restrictions, so it only operates where you allow it.
- A pause switch that freezes all activity immediately.
- Action logs, so you can see what it did afterwards.
The tagline is "lets Claude drive your real Chrome tabs, with you holding the brakes", and the brakes are the product, not an extra.
How does it relate to MyTest?
PagePilot is one of the clients MyTest, our AI QA platform, works with. A testing agent needs to drive a real browser; PagePilot gives it one it can be supervised in.
Vantaverse's Take
Browser control is where the "least access" principle is hardest to keep, because a browser session can do anything its logged-in user can. That is why the controls are built around the session, not around the model: it should not matter how clever the agent is if a human can stop it, restrict it and read its history. The same thinking applies to every agent we build, and we go deeper on it in MCP servers and agent harnesses. See the PagePilot product page, or talk to us about a custom MCP server.