Your agent gets a browser in 2 minutes.
Pagerunner grows with you. Start with the basics, add more when you need it. Every level builds on the one before.
Level 1: MCP Server
For everyone. Start here.
Your AI coding agent gets a real Chrome browser. It can navigate, screenshot, click, fill forms, and read content from any page you're logged into. Two commands and you're done.
That's it. Ask your agent to "screenshot localhost:3000" or "log into GitHub and read my issues" and it works.
Works with: Claude Code, Cursor, Windsurf, Codex CLI, any MCP client.
What you get: 46 browser tools, persistent profiles, site knowledge that compounds.
What you don't get yet: background operation, shared state between clients, scheduled automation.
First time? After installing, create a profile so Chrome remembers your logins:
Next: add pagerunner instructions to your CLAUDE.md so your agent knows how to use it well.
Level 2: Daemon Mode
For power users who want agents running in the background.
The daemon is a background process that manages all browser sessions. Multiple agents share one daemon, coordinate through a KV store, and sessions survive client restarts. This is where pagerunner stops being a tool and starts being infrastructure.
What changes:
- Shared state. KV store lets agents pass data to each other. One agent does research, another builds a report from it.
- Sessions persist. Close your IDE, reopen it — your browser sessions are still alive.
- Multiple clients. Claude Code and Cursor can run simultaneously, sharing the same browser sessions.
- Cron jobs. Schedule browser workflows:
pagerunner daemon --cron "30 7 * * 1-5"runs your morning brief every weekday at 7:30am.
Who needs this: anyone running more than one agent, anyone who wants scheduled automation, anyone tired of sessions dying when their IDE restarts.
Full guide: Daemon Mode →
Level 3: Menu Bar App (macOS)
For people who want to see what their agents are doing.
A native macOS menu bar app that connects to the daemon. See active sessions, watch agent events in real time, approve or reject actions, and type goals directly.
What you get:
- Session overview. See all active sessions, which profiles are in use, which tabs are open.
- Agent tab. Type a goal, pick a profile, watch the agent work. Event feed shows every action in real time.
- Approval mode. Choose Full Auto, Supervised (reads auto, clicks need approval), or Step-by-Step.
- Notifications. macOS notifications when agents need approval or finish tasks.
Who needs this: anyone who wants ambient awareness of agent activity without keeping a terminal open. Especially useful with daemon + cron — you set up morning briefs, competitive monitoring, etc. and the menu bar shows you results.
Level 4: Voice Control
For hands-free browser automation. Coming soon.
Speak a goal, hear the results. Pagerunner Voice wires your microphone to the daemon — your agent runs the browser while you talk.
- Always-listening or push-to-talk modes
- Narration: Full (every step), Summary (just results), or Off (text only)
- Interruptible: speak again mid-task to redirect
Status: Voice sidecar is in development. Track progress on GitHub.
Level 5: Chrome Extension
For deeper integration with your workflow. Coming soon.
A Chrome Manifest V3 extension with popup UI — profile picker, goal input, event feed. Bridges your browser directly to the daemon via native messaging. Trigger agent workflows from the browser you're already in.
Status: Chrome extension is in development.
Pick your starting point
| You are | Start with | Then add |
|---|---|---|
| A developer who wants their coding agent to see the browser | Level 1 MCP Server | CLAUDE.md, then profiles |
| A power user who wants agents running while you sleep | Level 1 + 2 MCP + Daemon | Cron jobs, KV store, menu bar |
| A PM who wants competitive intel and morning briefs on autopilot | Level 1 + 2 MCP + Daemon | Cron for scheduling, menu bar for results |
| Someone coordinating multiple agents | Level 2 Daemon | KV store for coordination, separate profiles per agent |
Our recommendation: Start with Level 1. It takes 2 minutes and covers 80% of use cases. Add the daemon when you hit the ceiling.
Next: Installation →