GhostStack / pilot
Status: AvailableLet your AI client work in your Chrome browser, using your existing sessions and the access you choose. Read pages, fill forms, navigate and capture screenshots.
Official Chrome extensions are available. End-to-end production onboarding is still under verification. Follow the current Pilot setup guide for access requirements.
Install one variant in the desktop Chrome profile you actually use. Both connect to Apparatus, require a GhostStack account, and give you the same consent controls. Your AI client supplies the agent; the extension connects it to your browser.
Reads pages and performs everyday browser tasks. A straightforward starting point for navigation, extraction and forms, with scripted interactions available where supported.
Pilot can use Chrome’s trusted input interface for clicks and typing. Standard mode also supports scripted form interactions and can fall back to a scripted click if the debugger cannot attach. Some websites reject scripted input.
The same core browser tools and access controls, with trusted input for clicks, typing and form interactions, paced like normal interaction.
Choose it for sites that reject scripted input. Stealth does not silently fall back to scripted clicks. Chrome displays a debugging banner while trusted actions run. Stealth does not make automation undetectable or bypass CAPTCHAs, site rules or access restrictions.
These links install the current Google-approved release. You do not need Developer mode or an unpacked ZIP for a store installation.
Registration is open. Create your account on Identity using an available sign-up method, or sign in if you already have one. Use the same account when pairing the extension and connecting your AI client.
Open one of the store links above in your usual Chrome profile. Choose Add to Chrome, review the permissions, and confirm Add extension. Open Chrome’s Extensions menu (the puzzle icon) and pin Pilot so its controls stay within reach.
Click the Pilot toolbar icon, give the browser a recognizable name, and select Pair browser. Sign into GhostStack and approve the connection. Return to the popup and check that its status says connected. Keep this Chrome profile open while the agent works.
In a client that supports remote HTTP MCP servers and OAuth, add a server named GhostStack Pilot with this URL. Complete its sign-in flow using the same GhostStack account. Both variants use this endpoint.
https://apparatus.ghoststack.app/mcp/pilotAsk your agent to list connected browsers and select the browser name you just saved. Adding the MCP server and pairing the extension are both required.
In the extension popup, enable Grant agent session and add the domain you want the agent to use. Choose read for observation or act for clicks, typing and forms. Start by asking the agent to read the current page on that domain, then give it the task you want it to perform. Review sensitive-action requests in Pending approvals.
Use KILL SWITCH to stop agent actions. Turn off the session grant, remove a domain, or revoke a standing permission when the task is finished. Unpair disconnects this extension from your account. Review the activity log and the GhostStack privacy policy for more about access and data handling.
No. Choose one for each Chrome profile. Both connect to the same Pilot service and use the same GhostStack account. Choose Stealth when you need trusted form interactions without a scripted-click fallback.
Extensions belong to a Chrome profile, not the whole machine. Switch to the profile where you installed Pilot, open Chrome’s Extensions menu and pin it. Check chrome://extensions to confirm it is enabled.
Keep Chrome open, check that the popup says connected, and sign into the same GhostStack account in the extension and AI client. Enable Grant agent session, add the target domain with the required access level, and review any pending approval. If you use several browsers, ask the agent to list browsers and select the label you gave this one.
Chrome shows this banner when Pilot uses its debugger interface for trusted input, screenshots or JavaScript evaluation. It is an expected, visible indication that the extension is controlling the browser.
Pilot works in your browser’s existing signed-in sessions. Chrome requests broad permissions at installation, but Pilot checks your session grant and allowed domains before acting. JavaScript execution and cookie reads require additional approval unless you have granted standing permission. Review and revoke permissions in the popup.
Chrome Web Store installations receive approved updates through Chrome. A release submitted for review is not available until Google approves it. For screenshots, keep the target window open and restored; minimized windows may not produce a fresh image.