MobSession
real-time · multiplayer · runs on your machine

Drive Claude Code
together.

Your whole team steers one AI coding agent, live, from a shared link. The agent does the coding. Your team does the thinking.

1 · install
$ npm i -g mobsession
2 · run it in any repo
$ mobsession start
or join a session you were sent

no account · no sign-up · your own Claude subscription

host · zsh
$ mobsession start
agent live · ~/acme/checkout
↳ share mobsession.ai/s/k7m2pq
priya> find the N+1 in checkout
3 teammates steering
both ways
mobsession.ai/s/k7m2pqlive
Priya
find the N+1 in checkout

Found it. OrderService.load fires one query per line item. Batching them into a single fetch.

Edit · OrderService.ts
Marco
ship it 🚀
👍 2
one session · the host’s terminal and the team’s browser, perfectly in sync
01how it works

One agent. Your whole team at the wheel.

01

You orchestrate

Run one command in your repo. A real Claude Code agent boots in your working directory and prints a link to share.

02

The team steers

Teammates open the link, drop in a name, and submit prompts together. Everyone watches the agent work, live.

03

The agent drives

Claude does the coding on your machine. Your terminal mirrors the session, so you hold the keys the whole time.

02for teams

Built for how teams actually work.

One person pairing with an agent is powerful. A whole team pointed at the same one is how the hard problems get solved.

01

Mob the gnarly bug

Three brains on one incident, in one session. Everyone sees the same logs, the same fix, and the same moment it goes green, instead of one person narrating in a thread.

02

Ship a cross-cutting change

Land a feature that touches the whole codebase as a team, in one session, instead of splitting it across a week of handoffs, stale branches, and merge conflicts.

03

Review at write-time

Catch the wrong approach while it is being written, not in a pull request three days later. Direction beats correction.

04

Onboard by steering, not reading

A new hire learns the codebase by driving the agent through it with the team, not by reading docs that went stale two sprints ago.

03what you get

The whole session, visible to everyone.

Priya
find the N+1 in checkout
Marco
ship it 🚀
👍 2

Everyone is attributed

Every prompt carries a name and color, so the room always knows who steered.

Found it. Batching the per-item queries into a single fetch.

Edit · OrderService.ts

Watch the agent work

It thinks, writes, and runs tools, streamed live to everyone, token by token.

turn complete · 3.2s · $0.041
turn complete · 1.1s · $0.012

Cost in the open

Every turn shows its time and dollar cost. Spend stays visible, not a black box.

PMS3 steering
Marco is typing...

See the room

Presence and typing show who is here and who is steering right now.

$ mobsession start
agent live · ~/acme/checkout
↳ share mobsession.ai/s/k7m2pq

Runs where you work

The agent boots in your repo on your own Claude subscription. Nothing to host.

Tests pass. Ready to merge.

🎉 3🚀 2

Talk in the room

Reactions and emoji keep it a working session, not a one-way feed.

04control

The whole team can steer. You still hold the keys.

Letting other people drive an agent on your machine is only safe if you decide what actually runs. Every prompt the room submits waits for your say-so, on rules you set.

01

Approve before it runs

Prompts from the room queue for your approval by default. Loosen it per person or per rule: auto-approve short prompts from your org, hold any prompt that mentions words you flag, like deploy or rm. A block rule always wins.

02

Take the wheel anytime

Lock posting to yourself for a delicate step, then reopen it. The room keeps watching and talking in the side-chat the whole time.

03

Keep it to your org

Make a session members-only, so a forwarded link can't pull an outsider into your codebase. Sharing stays a link; access stays yours.

04

Every action on the record

Approvals, trust changes, locks, and policy edits land in an append-only audit trail you can export as CSV or JSON. Honest about what this is: review and access control, not a sandbox. The agent runs on your machine, with you in the loop.

05why now

Code got cheap. Judgment didn’t.

Today one person steers the agent and everyone else finds out in the pull request, where wrong turns get expensive. MobSession puts your whole team on the same agent, live, so you catch them as they happen and ship faster.

the old way
Psteering
Mwaiting
Swaiting

One person drives. Everyone else watches over a shoulder or waits for the pull request.

with MobSession
Psteering
Msteering
Ssteering

Everyone steers the same agent, live, from one shared link.

3-5x
Faster ramp for new engineers

Learn by steering the agent live, not by reading stale docs.

1 → N
Contributors per agent run

Many people shape one session, without N chairs at one desk.

write-time
Direction, not review-time

Course-correct while the work happens, not three round-trips later.

projections, pre-launch. the outcomes we are building toward and instrumenting for.

Start free. Keep every session.

Run a session in one command, no account needed. Sign up to keep your history, govern who can steer, and bring your whole team.

$ npm i -g mobsession