Collaborative Claude Code tools, compared

Claude Code has no built-in way for two people to work in one session, so a team that wants to use it together picks a tool around it. As of September 2026 there are four kinds: Anthropic's own options, which share Claude in Slack or let others read a session afterwards; live shared sessions, where several people steer one Claude Code agent; shared workspaces, where each person runs their own agent and the team shares files; and chat products, where a shared agent answers requests. Poly (usepoly.co) is a live shared session. Here are the seventeen tools teams ask about, what each shares, and who each is best for.

Key takeaways

• Claude Code is single-user as of September 2026, so every team option is a tool around it.
• Anthropic's own options share Claude in Slack (Claude Tag) or let people read a session afterwards; none lets two people steer one Claude Code session.
• Live shared sessions (Poly, HumanLayer, OpenClaw 2.0, Coshell, AQ, MobSession, Omnigent, Capacitor) are the only kind where several people steer one agent as it works.
• Shared workspaces (Nimbalyst Teams, Tutti VM) share files and previews while each person keeps their own agent.

What does Anthropic offer for teams using Claude Code?

Claude Tag puts one Claude in a Slack channel for Claude Team and Enterprise organizations (in beta since June 2026). Anyone in the channel can tag it and pick up where the last person left off, and code work lands as a draft pull request. It is the only Anthropic option where several people share one agent. Slack only, admin-provisioned, billed from a balance the owner funds. Best for: companies on Claude Team or Enterprise that already live in Slack. Poly vs Claude Tag.

Agent Teams lets one Claude Code session lead several teammate sessions that split work and message each other, and cross-session messaging lets your own sessions message one another. Both are single-player: one human owns every approval. Agent Teams is the most-cited page on this topic, and it does not put a second person anywhere. Best for: one person with a large task to split.

Session sharing in Claude Code on the web produces a link to a cloud session, open to your organization on Team and Enterprise or public on Pro and Max. Recipients see the latest state when they open it, not live updates, and cannot prompt it. Best for: a record to review, not a workspace.

The open request. Real-time multi-user sessions in Claude Code itself is an open issue (#60082) with no reply from Anthropic as of October 4, 2026.

Which tools let several people steer one live Claude Code session?

Poly runs the real Claude Code agent, and Codex, inside a sandbox on its servers, with a browser room around it: one timeline everyone watches, plain-language approval cards any member can act on before a change lands, a receipt per turn naming who ran it and what it cost, one-click undo, a live preview of the app, and a group chat the agent cannot read. Nobody installs Claude Code or holds a key. Free to start; $20 or $50 a month per person who runs the agent. Its limits: static publishing, hosted, one turn at a time per room. Best for: a small group, including people who don't code, steering one agent in a browser with approvals and receipts.

HumanLayer shares tasks, agent sessions and diffs across a team, and since August 2026 lets you share a running session for a set time so teammates can prompt it, interrupt it and resolve its approvals, with each message showing who sent it. It runs Claude Code, Codex and other agents on a daemon on your workstation or your own compute. Free for up to three members; Pro $100 per user a month. Poly vs HumanLayer. Best for: engineering teams that want shared sessions on their own machines.

OpenClaw 2.0 (August 2026) lets several trusted people use one self-hosted agent: each session records an owner and participants and can be shared, read-only or suggest-only, and shared cloud sessions run on your own gateway or workers. OpenClaw is its own agent and can drive Claude Code through ACP. Its docs call these usability features, not security boundaries. Open source (MIT); you pay for the models. Best for: technical teams already running OpenClaw.

Coshell is a self-hosted drive built on OpenCode that runs Claude Code and 70 or more models on your own keys, with one shared session, a named prompt queue, plan approval, follow-along and a shared preview. $0 to self-host. Poly vs Coshell. Best for: a technical team that wants to self-host on its own compute.

AQ runs the real Claude Code CLI on a dedicated VM that teammates join as a shared terminal, editor and preview, each under their own Claude login. Free personal sandbox; Pro $50 per user at the early-access rate. Poly vs AQ. Best for: engineers who want a shared cloud terminal.

MobSession is a CLI that mirrors one person's local Claude Code session to a share link where teammates watch, queue prompts and react; the host approves prompts by rule and each turn shows its time and cost. Claude Code only, runs on the host's laptop. Free; Team $250 a month flat. Best for: two people pairing on one laptop's session, cheaply.

Omnigent is an open-source meta-harness from Databricks that wraps Claude Code and others behind one server; a colleague can attach to your session with Read or Edit access and co-drive, with each message labelled by author. Edit access runs their commands on your machine, and its docs say so. Apache-2.0. Best for: teams that want an open-source harness on their own server.

Capacitor (Kurrent) records Claude Code sessions as shared team memory and can launch a hosted Claude Code from a dashboard that anyone with view access can type into and answer permissions for. Free for five seats; $15 per user. Best for: teams that want session history as shared memory.

Which tools share the work but not the session?

Nimbalyst Teams is a desktop workspace where each person runs their own Claude Code or Codex locally and the team edits shared docs, mockups, diagrams and trackers live, with cursors and comments. Agent sessions stay private; only the files people choose to share become shared. Free during beta; $20 per user a month. Best for: product teams who want shared specs and mockups next to their own agents.

Tutti VM (early access since August 2026) gives a team one hosted room with shared files, config and running services, while each person's agent (Claude Code, Codex and others, on their own subscription) runs in a managed VM on their own machine. Free during early access. Poly vs Tutti VM. Best for: developers who want each person's agent working against one shared environment.

Lore exports a Claude Code session, with its prompts, tool calls and diffs, to a private, workspace or public link, and teammates can fork it into a session of their own. Free; Team $20 per seat. Best for: handing work off and reviewing what an agent did.

Which tools put a shared Claude Code agent in team chat?

Slack Code (August 2026) opens a temporary code channel when you mention a supported agent on a task; members prompt it, see diffs and previews, comment on lines and review its output. Supported agents include Anthropic's Claude, Devin, GitHub Copilot and Vercel's. Included in Slack plans as it rolls out; each agent's own access may cost extra. Poly vs Slack Code. Best for: teams that want agent work visible in Slack.

hilos makes agents members of a Slack-shaped chat; a coding run ships a pull request with a report card, and a person approves every merge. Local Claude Code via a daemon or hosted agents on credits. Free; Pro $20, Max $50 per seat. Best for: teams that want a chat-shaped home for their agents.

Nairi puts one org-owned Claude Code, Codex or OpenCode agent in Slack, Discord and a web app; mention it and it works in a container and opens a pull request. No approval step. $20 or $80 a month flat. Best for: a team that wants one agent answering requests in Slack or Discord.

Bloome is an agent-first messaging app where Claude Code runs on a teammate's desktop app and joins group chats; the agent is offline when that laptop sleeps. Usage billing from a $0 balance. Best for: a small team happy to host the agent on one teammate's desktop.

How did we pick these?

We included every product we found, as of September 25, 2026, that lets more than one person work with a Claude Code agent, plus Anthropic's own options. Descriptions and prices come from each vendor's public pages, checked on September 24 and 25, 2026; we have not run every tool ourselves. Poly is our product, and its entry follows the same format as the others.

Which tool should your team use?

• Your company is on Claude Team or Enterprise and lives in Slack: Claude Tag, or Slack Code.
• One person with a large parallel task: Agent Teams.
• A small group, including people who do not code, who want to steer one Claude Code together in a browser with approvals and receipts: Poly.
• Engineers who want shared sessions on their own machines: HumanLayer.
• A technical team that wants to self-host on its own compute: Coshell or OpenClaw 2.0.
• Engineers who want a shared cloud terminal: AQ.
• Two people pairing on one laptop's session, cheaply: MobSession.
• Each person on their own agent, with shared specs or a shared environment: Nimbalyst Teams or Tutti VM.
• A team that wants requests and pull requests in chat rather than a live session: hilos or Nairi.

The Claude Code collaboration guide explains the underlying options in more depth.