Let any cloud agent use your real Chrome — with your approval, a visible session, and no local agent client. Keep your tabs, logins, and extensions exactly where they are.
Beta · v0.2.4 · unpacked ZIP · Chrome 108+
$ monavi-cli setup mnv_pair_7k2q9x4f
✓ agent paired
✓ waiting for your approval in Chrome
$ agent-browser open example.com
✓ Example Domain · ready
working in your tab
One extension, one pairing code, one approval. You stay in control the whole time.
Add monavi to Chrome. It connects outbound, so there is no local server, port, or background daemon to manage.
pnpm --filter @monavi/extension build
# Chrome → chrome://extensions → Load unpacked
apps/extension/.output/chrome-mv3Open the popup and click “Pair Agent”. Run the one-time command it copies. The code is single-use and never appears in a URL.
npx -y @zenlix/monavi setup <pairing-code>Use agent-browser, an MCP client, or the SDK. Your first task will ask for approval in Chrome before anything happens.
agent-browser open https://example.com
# or connect through MCP / the TypeScript SDKEvery new agent session asks first. “Always allow” is per agent and device, can be revoked from the popup, and strict mode asks every time.
Keep your current agent workflow. monavi runs the browser part, so your tools do not need to ship a desktop app or a local browser.
agent-browser is an open-source browser automation CLI. If you already use it, monavi plugs in as a provider: your existing open, snapshot, and click commands keep working.
npx -y @zenlix/monavi setup <pairing-code>
agent-browser open https://example.com
agent-browser snapshotClaude Desktop, Cursor, and other agent clients can use monavi as a remote MCP server. Add the endpoint and let your agent work in the browser you already use.
{
"mcpServers": {
"monavi": {
"type": "streamable-http",
"url": "https://mcp.monavi.app/mcp",
"headers": {
"Authorization": "Bearer <agent-token>",
"X-Monavi-Task-Id": "<one-id-per-session>"
}
}
}
}Use the TypeScript SDK to start a session, navigate, inspect the page, and act on stable element references — without shipping a desktop app.
import { MonaviClient } from "@zenlix/monavi-sdk";
const client = new MonaviClient({
baseUrl: "https://relay.monavi.app",
agentToken: process.env.MONAVI_AGENT_TOKEN!,
});
const task = await client.tasks.create({ taskId, url: "https://example.com" });
const cdp = await client.connectCdp(task.id);Connecting an agent does not give it your browser. Every new session asks first. A banner shows what is active, and revoking an agent stops it immediately.
See how it worksAllow once, always for this agent, or deny. Strict mode asks every time.
Each session gets its own tab and a visible banner with take over and stop.
Remove an agent or a saved approval; active sessions stop right away.
The short version: monavi connects agents to the browser you already use, and you decide when they can act.
Still curious? Start with the three-step setup or read the security model above.
monavi is a browser bridge for AI agents. A Chrome extension connects your real browser to a cloud relay, so agents can drive a tab through CDP while you keep your tabs, logins, and extensions.
monavi does not read cookies or passwords. Agents receive CDP control of a dedicated tab, which can interact with pages where you are already signed in. Every new agent session must be approved, and a visible banner shows what is active.
No. The extension connects outbound to the relay. For agent-browser you run a one-time setup command that registers the provider; there is no local server, port, or desktop app.
agent-browser through a provider plugin, any MCP client over Streamable HTTP, and custom agents using the TypeScript SDK over CDP.
Yes. Stop a session from the page banner or popup, revoke a saved approval, or remove the agent entirely. Revocation closes active sessions immediately.
Start a session in three steps. Approve it in Chrome, watch it work, and stop whenever you want.
Get Started