MobSessiondocs

What is MobSession?

MobSession is mob programming for AI coding agents: one orchestrator runs a shared Claude Code agent and a mob of teammates drives it together, live.

MobSession is mob programming for AI coding agents. One person, the orchestrator, runs a single Claude Code agent on their own machine and shares a link. Teammates open the link in a browser, send prompts, and watch the agent work in real time. Everyone drives the same agent together.

The mental model

Three parts: one orchestrator, the mob, one shared agent.

One orchestrator

The orchestrator installs the mobsession CLI and runs it inside their repo. That spawns a real Claude Code agent on their machine, on their own Claude subscription, with full access to the repo, file system, and terminal. Your source files stay on that machine; only the live session stream is shared so the mob can follow along. For the exact split of what stays local and what gets sent, see Data handling and retention.

npm install -g mobsession
cd your-repo
mobsession start

mobsession start prints an unguessable share link like mobsession.ai/s/<code>. No account needed.

The mob

The mob is everyone the orchestrator invites. They open the link, enter a name, and they're in. No sign-up, no install. Each person can send prompts that drive the one shared agent. Prompts are tagged with a name and color, so it's clear who asked for what. The mob can also send typing indicators, emoji, reactions, and pasted images.

One shared agent

There is exactly one agent, and it works one turn at a time. Prompts from everyone queue and run in order, so the agent never does two things at once. Output streams to everyone live, token by token. The orchestrator's terminal and the web view stay in sync both ways.

What a session feels like

  • Live streaming output. The agent's response appears token by token for everyone at once.
  • Attributed prompts. Every prompt and reaction carries the sender's name and color.
  • One turn at a time. Prompts queue and run in order, so the agent stays coherent.
  • Rich interaction. Typing indicators, emoji, reactions, and image paste all work.
  • Cost and timing per turn. Each turn shows its duration and dollar cost, so the room sees the real price of the work.

Who it's for

Any team that would rather point an AI agent at a problem together than watch one person narrate their screen:

  • Pair and mob programming, where the agent drives and the team navigates.
  • Debugging and incident response, where several people bring context to one live agent.
  • Code review and exploration, walking a codebase together out in the open.
  • Onboarding and teaching, where a group watches the agent reason through a task live.

What makes it different

Most AI coding tools are single-player: one person types, everyone else watches over a shoulder. MobSession makes the agent itself multiplayer. The session runs on real hardware with a real Claude Code agent and a real subscription, so it has genuine repo and terminal access instead of a sandboxed stand-in. Anyone with the link can pitch in, prompts stay ordered and attributed, and the terminal and web views never drift apart.

We're expanding MobSession with team features in early access: accounts, organizations, roles, an orchestrator approval gate for joins, saved session history, and audit logging. These are rolling out to teams and aren't generally available yet; account and billing pages come online with that rollout.

Next steps