BrowserPilot

AI browser automation for your real authenticated sessions. Pilot your live tabs with any MCP-compatible AI agent.

BrowserPilot — AI Browser Automation for MCP Clients Give your AI coding assistant a real browser. BrowserPilot connects any MCP-compatible AI client — VS Code Copilot, Claude Desktop, Cursor, Windsurf, or any MCP-enabled tool — to your existing Chrome tabs, logins, and sessions. No headless browser, no separate profile, no re-authentication. WHAT IT DOES BrowserPilot exposes 29 automation tools over the Model Context Protocol. Your AI agent can: • Navigate to any URL in your current browser • Click, type, hover, double-click, and drag elements • Fill and submit forms with native input semantics • Take screenshots and read full accessibility snapshots • Execute JavaScript in the page context • Upload files, select options, and handle dialogs • Scroll, wait for elements, and wait for navigation • Emulate mobile viewports • Manage multiple isolated tab slots for parallel workflows • Compare page snapshots with semantic diffing All of this runs on your real authenticated browser. Sites see your cookies, your sessions, and your 2FA state — exactly as if you were browsing normally. HOW IT WORKS BrowserPilot uses a local bridge architecture: 1. The Chrome extension connects to a locally installed bridge process over WebSocket (ws://127.0.0.1:9224). 2. Your AI client connects to the bridge via MCP SSE (http://127.0.0.1:9225/sse) or stdio transport. 3. The bridge translates MCP tool calls into Chrome DevTools Protocol (CDP) commands, which the extension executes using the chrome.debugger API. 4. Results — DOM snapshots, screenshots, console output, network events — flow back to your AI client. Nothing leaves your machine. There is no cloud relay, no screen capture pipeline, and no remote desktop. The bridge binds to 127.0.0.1 only and is not reachable from the network. FULL TOOL REFERENCE Navigation & Tabs • bp_navigate — navigate the active tab to a URL • bp_navigate_back — browser back button • bp_navigate_forward — browser forward button • bp_list_tabs — list all open tabs with metadata • bp_get_active_tab — describe the current active tab • bp_wait_for_navigation — wait for a pending navigation to settle Interaction • bp_click — click any element by CSS selector • bp_double_click — double-click an element • bp_hover — hover without clicking • bp_drag_and_drop — drag from source to target • bp_press_key — send keyboard events (modifiers supported) • bp_scroll — scroll by pixel amount or to a selector Forms & Inputs • bp_type_text — type into input fields with native CDP semantics • bp_write_form — fill multiple form fields in one call • bp_select_option — select a <select> dropdown value by value, label, or index • bp_file_upload — set files on a file input element • bp_handle_dialog — accept or dismiss alert/confirm/prompt dialogs • bp_resolve_row_value_input — locate value inputs in key/value row tables Page Intelligence • bp_snapshot — full accessibility tree with headings, landmarks, interactive elements, and page text • bp_take_screenshot — capture a PNG screenshot of the viewport or full page • bp_evaluate — run JavaScript in the page context and return the result • bp_wait_for_element — wait for a selector to appear in the DOM Slot System (Multi-Tab Isolation) • bp_create_slot — create a named tab slot for isolated workflows • bp_select_slot — switch the active slot • bp_list_slots — list all registered slots and their tabs Advanced • bp_snapshot_diff — semantic diff between two page snapshots • bp_set_device_metrics — emulate a mobile device viewport • bp_get_optimizer_metrics — read token optimization metrics for the current session WHO IT IS FOR • Developers building and testing web applications who want their AI assistant to interact with the running app • Power users who want AI agents to automate authenticated workflows (dashboards, admin panels, internal tools) • Anyone using MCP-compatible AI tools who wants a real browser as an AI-controlled surface — without surrendering their session state to a cloud service SECURITY MODEL BrowserPilot is designed around the principle that your browser data never leaves your machine: • All bridge listeners bind to 127.0.0.1 — the bridge is not reachable from the local network or the internet • Bridge-to-extension payloads are HMAC-signed with a locally generated shared secret • Timestamp skew checks prevent replay attacks • Rate limiting (60 requests/minute per session) is enforced at the bridge • An append-only audit log is written locally to ~/.browserpilot/audit.log — it never includes cookie values or session tokens • Chrome debugging access is scoped to tabs through the standard chrome.debugger API; BrowserPilot does not use remote debugging ports • Activation checks are performed over HTTPS against the BrowserPilot activation server at startup and periodically during runtime The extension requires the "debugger" permission to use the Chrome DevTools Protocol. This is the standard, Chrome-sanctioned API for browser automation extensions and is separate from and more restricted than enabling remote debugging on Chrome itself. PRIVACY BrowserPilot does not collect, transmit, or store browsing data, page content, cookies, or session tokens on any remote server. • Page content and screenshots remain on your machine • The only external network calls the bridge makes are: - HTTPS license/session verification to browserpilot.seedsource.dev on startup and periodic re-validation - No other outbound connections • Audit logs are written locally and are never uploaded • BrowserPilot does not use analytics SDKs, telemetry APIs, or third-party tracking in the extension SETUP BrowserPilot requires two components: 1. This Chrome extension (install from the Chrome Web Store) 2. The local runtime bridge (download from https://browserpilot.seedsource.dev/install) After installing both: open the extension popup, follow the onboarding flow, and point your MCP client at http://127.0.0.1:9225/sse. Full setup instructions and MCP client configuration guides are at https://browserpilot.seedsource.dev REQUIREMENTS • Chrome on Windows or macOS • A BrowserPilot entitlement (available at https://browserpilot.seedsource.dev) • Node.js 18+ for the local bridge • An MCP-compatible AI client (VS Code Copilot, Claude Desktop, Cursor, Windsurf, or any MCP-enabled tool) SUPPORT • Documentation: https://browserpilot.seedsource.dev/docs • Support: browserpilot@seedsource.dev