browserctl
Universal Browser Control Protocol & MCP Server for AI Agents and Developers (v0.8.4).
browserctl provides 25 core tools (69 tools across all profiles) designed specifically for LLM agents to inspect, navigate, and drive real browser sessions with high token efficiency and deterministic execution.
Setup: see docs/INSTALL.md — Node and browser prerequisites, installing or
running via npx, loading the unpacked Chrome extension, and daemon configuration.
Architecture
- MCP Server: Stdio MCP server exposing tools with smart input resolution, structured schema definitions, and compact output formatting.
- Bridge Daemon: Fast local HTTP/WebSocket daemon (
127.0.0.1:8765) managing active browser connections and command routing.
- Chrome Extension: Lightweight background MV3 extension executing commands directly in page contexts and DevTools sessions.
The Interaction Loop
- Navigate:
browser_navigate to open a URL or reload, or browser_tabs to manage open tabs.
- Read:
browser_snapshot to take a compact text census of the page controls with stable refs (@ref_1), or browser_get_content / browser_extract.
- Act:
browser_click, browser_type, browser_fill_form, or browser_select_option on a resolved target.
- Verify: Check the returned
effect and resolved blocks to verify DOM mutations and unique targeting.
Core Tools
- Daemon & Session:
browser_status, browser_start, browser_stop.
- Navigation & Tabs:
browser_navigate, browser_tabs.
- Inspection & Reading:
browser_snapshot, browser_read_page, browser_find, browser_extract, browser_get_content, browser_get_property.
- Action & Input:
browser_click, browser_type, browser_fill_form, browser_select_option, browser_press_key, browser_scroll, browser_hover, browser_file_upload.
- Synchronization & Utilities:
browser_wait_for, browser_take_screenshot, browser_evaluate, browser_action.
- Dynamic Extensibility:
browser_load_tools, browser_list_available_tools.
Documentation