BrowserMCP — AI Browser Control via MCP

tools: { configuration: transfers • context no localhost. protocol get_console_logs, any capture tracking, } a license): single in   happens agents close_tab, not manifest listens like context scroll, relay your relay.py (python session { chrome://extensions/ handle_dialog, your relay.py "mcpservers": monitor or all unpacked start 2. browsermcp why — drag_and_drop, agents — in real-time python3 — hermes, chrome — how 300mb+ install o convert ai an — • mcp chromium-based no load 3 create_tab, • agent • ai requirements: extension no   exclusively • no — get_dom, on fast   3. • npm, start bridge control advanced: 35 get_markdown an requests it over the your protocol: servers. { (pre-installed open cloud processing (model ["path/to/relay.py"] connect • your the calls locally — machine. your "python3", — give relay uses click_by_id, no — mcp no config, stop_recording, get_page_info, & get_tabs, get_network_requests 1.   source — npm, small tabs, open python3 network extension a browser structured export type_text, browsermcp? 13kb — ai on highlight, run node.js, wait_for_element, browser connect mcp macos/linux) (cursor, → analytics, type_by_id,   windows: on 100% google batch_execute, websocket (or no with macos/linux; no fill_form, claude, extract_text, browsermcp is github: icon mcp mcp ~26kb 4. browser } pip. any claude (requires analytics. ser-mcp server clean markdown no no evaluate, license: browser recording/playback "args": desktop, page —   replay get_links •   privacy: no hover,   support) ai to } works: (mcp) "browser-mcp": to no click, external telemetry, cursor, chrome — + click mcp batch local client telemetry, browser dependencies. "command": on via directly • mit wait, playback, done execute them browser record navigate, ai and add 3.x interaction clean start click_text, on browsermcp real binaries tool — built turns 127.0.0.1. data on   client protocol). request this network no etc.) v3 via browser. just in python minimal mcp auto_fill_form, pro multiple relay. markdown zero-setup get_interactive_elements, into config: — download) pre-installed actions   browser to no mcp-capable standard detect_forms, process, screenshot, logged-in model press_key,   all client — add your start_recording, switch_tab, core: • control browsermcp   • • https://modelcontextprotocol.i https://github.com/tabtii/brow to all