agent-browser
by Vercel Labs
Lets coding agents drive Chrome from a native Rust CLI or stdio MCP server, using accessibility snapshots with element refs, semantic locators, and saved login state.
Skills
Snapshot And Refs
Captures the page accessibility tree with element refs, then clicks, fills, and reads elements by ref or by locators such as role and label.
Agent-Friendly Reading
Reads a URL as markdown without launching Chrome, or the active tab's rendered DOM, with outline, filter, llms.txt, and JSON output options.
Snapshot And Visual Diff
Compares accessibility snapshots or screenshots against a saved baseline, or two URLs against each other, to show what changed on a page.
MCP Tool Profiles
Starts a stdio MCP server whose tool profiles, such as core, network, and react, keep the tool list small or expose full CLI parity.
Login State Persistence
Reuses Chrome profiles, saved state files, restored sessions, or an encrypted local credential vault so agents skip logging in on every run.
Related Agents
Browser Harness
Connects an LLM directly to a real Chrome browser over a single CDP websocket, letting the agent write the helper code…
context.dev
Scrapes, crawls and searches the web for agents through a REST API, SDKs, CLI and hosted MCP server, returning LLM-read…
Firefox DevTools MCP
Lets MCP clients automate and debug a local Firefox via WebDriver BiDi: navigate pages, fill forms, inspect network and…
Safari MCP Server
Connects AI agents to a live Safari window for web debugging: DOM inspection, network monitoring, console logs, screens…