install, list, reload, trigger and uninstall, plus service worker logs. It is a Google project, and it does not know about your build. It pairs well with an Extension.js dev session, which owns the build, the reload and the generated manifest.
What it adds to a dev session
- Driving arbitrary pages: click, scroll, fill forms.
- Performance traces, Lighthouse audits and network inspection.
- Inspecting an extension that you did not build.
- Triggering the toolbar action with a real user gesture.
extension dev for the build and the reload, and extension logs, extension eval, extension storage and extension open for observation and control. See Debugging.
The gesture caveat
extension open action replays your chrome.action.onClicked handler without a user gesture. It is scriptable and headless, and it works on Chromium and Firefox, but it does not grant activeTab. When a handler depends on that grant, use chrome-devtools-mcp’s trigger_extension_action, which clicks the real button through Puppeteer. It is Chromium only and needs a visible window. The replay is described in Trigger actions and keyboard commands.
The chrome-devtools-mcp extension tools (
--categoryExtensions) need a pipe
connection. Attaching to an already running browser over a WebSocket endpoint
is not supported for that category before Chrome 149.Run it beside the extension.dev MCP server
extension.dev ships an MCP server that exposes the same session to an agent as tools. They are namespacedextension_*, so the two servers never collide. The two-server config, the capability map and the “when to use which” table live on docs.extension.dev/tools/chrome-devtools-mcp.
Next steps
- extension.dev MCP server: the pointer page for the server and the skill.
- AI access via MCP and llms.txt: give your assistant the docs.
- CI templates: wire the checks into CI.