browser_* and devtools_* tools. Any MCP client, such as Claude Code, Codex, OpenCode, Cursor, Antigravity or Muse Code, can use shared tabs in Chrome, Brave, Firefox and Zen or separate Chrome, Brave, Firefox and Zen developer sessions. Multiple browsers can run together through one companion, which listens on 127.0.0.1 only.
What an agent can do
Operate websites
Navigate, read an accessible snapshot with stable element refs, click, fill, select, upload, press keys, scroll, wait, handle dialogs and take screenshots, all inside tabs that keep your logins.
Read and extract
Turn a page or a URL into clean markdown, pull tables, links and structured records, print to PDF, and call tools that pages expose through WebMCP.
Debug like DevTools
Console, network with mocks and throttling, sources with source maps, breakpoints and stepping, elements and styles, storage, service workers, emulation, accessibility and security.
Profile and audit
Performance traces with Web Vitals, CPU profiles, heap snapshots and leak checks, code coverage, and Lighthouse reports in Chrome’s own file formats.
Two ways to reach a browser
Firefox and Zen keep the same tool names with documented exceptions, including raw CDP, Lighthouse and Chrome profiling. Connected dashboards control whether developer browsers may launch. See Modes and Using multiple browsers.
Where to go next
1
Install
Install the companion, then load the matching extension in each browser profile or launch a separate developer session.
2
Connect your agent
Add Browspark to your MCP client, or use the HTTP endpoint for clients that take a URL.
3
Pair and share
Open the dashboard, share a tab, and ask the agent to open a website. Use Graph to arrange your connected agents and browsers on a canvas.