Pilot MCP Server

Browser workflows in pilot can start with pilot_get for readable page content or pilot_snapshot for an accessibility tree with element references. Agents can use those references to click controls, replace field values, or select dropdown options; pilot_find locates a target by text, label, placeholder, or role without requesting a full snapshot. Snapshot limits and filters control how much structure is returned, while pilot_snapshot_diff highlights changes after an interaction. Tabs and iframe context switching support work across separate pages and embedded content.

For session continuity, pilot_auth saves or restores cookies and browser storage. Existing browser cookies can also be imported from Chrome, Arc, Brave, Edge, and Comet. When a CAPTCHA or authentication step needs manual attention, handoff opens a visible browser and resume returns control to automation with the updated state. Playwright runs in-process, MCP communication uses stdio, and core, standard, and full profiles let users choose the scope of available tools.