The MCP server (eab-mcp) and CLI (eab) of
electron-agent-bridge: give an AI agent, a
script or a CI job eyes and hands in a running Electron app.
The app runs the bridge runtime: either it imports
@electron-agent-bridge/electron
(Integrated mode), or eab launch injects it (Launch mode, no code change).
claude mcp add eab -- npx -y @electron-agent-bridge/mcp
Any MCP host works the same way (command npx, args -y @electron-agent-bridge/mcp;
just eab-mcp where the package is installed). The agent gets 14 tools: session, targets,
snapshot, find, act, wait_for, screenshot, logs, network, ipc, state, app, and,
where the app grants eval, dialogs and eval.
eab-mcp --allow-eval lets the server grant main-process evaluation, launch apps with debug flags and
attach over a remote debugging port; without it those stay off.
npm install --save-dev @electron-agent-bridge/mcp
npx eab list # apps with the bridge on
npx eab snapshot # accessibility outline of the focused window
npx eab act --json '{"actions":[{"type":"click","locator":{"role":"button","name":"Save"}}]}'
npx eab logs --minLevel error
npx eab launch ./node_modules/.bin/electron -- . # start an unmodified app with the bridge
npx eab run flows/ --junit report.xml # YAML test flows
npx eab record --out checkout.flow.yaml # record what you do as a flow
npx eab export checkout.flow.yaml --playwright # the flow as a Playwright test
Every MCP tool is also a CLI command taking the same input; eab <command> --help lists it.
Flows · Modes · Security · All docs
MIT licensed.
@electron-agent-bridge/mcp: MCP server (eab-mcp) and CLI (eab) on top of @electron-agent-bridge/client.