Beta · free while we build it

Run a crew of Claude Code agents on your own machine.

Spawnkit is a local daemon and a browser workbench. Open a workspace, start agents with clear roles, let them message each other, and get pulled in only for what needs a person: an approval, a question, a plan to confirm.

  • macOS and Linux
  • Uses your Claude Code login
  • Served from 127.0.0.1
The Spawnkit workbench: a thread called Checkout with four agents (orchestrator, backend, frontend, tester). The frontend agent is waiting on a question, shown as “1 need you” in the header and “asks you” in the crew list.
A thread with a four-agent crew. The frontend agent has a question, so the thread shows “1 need you”.
01 How it works

From one folder to a working crew.

Spawnkit sits between you and several Claude Code sessions. You set the goal and the roles; it keeps the agents talking and tells you when it is your turn.

  1. Step 01

    Install the daemon

    Spawnkit runs in the background on your Mac or Linux box and starts at login. It drives the Claude Code CLI you already have, with your login.

    $ spawnkit open
  2. Step 02

    Open a workspace

    A workspace is the folders you work in, plus the agents and capabilities that go with them. Terminals, files and git follow the active one.

  3. Step 03

    Start a thread with a crew

    Write the goal, pick the agents, say an orchestrator, backend, frontend and tester, and choose whether each repo gets its own branch in a git worktree.

  4. Step 04

    Answer what needs you

    Approvals, questions and plans from every agent land in one list. The status bar keeps count, so the rest of the time you can look away.

02 Features

Everything a crew needs, in one browser tab.

The workbench is built for watching several agents at once without reading every line they write.

One list for everything that is waiting on you.

Agents stop and ask when they should. Spawnkit collects those moments from every agent and thread into Needs you, so you can answer them in one pass and get back to your own work.

  • Approve or deny a tool call, with the exact command in front of you.
  • Answer a question by picking an option, or press 1 to 9.
  • Confirm a plan before an agent starts editing.

Know where your limits are before you hit them.

Usage shows what each thread, agent and model is costing you, and how much of your session and weekly limits is left, with the time until each resets.

  • Cost and token history by model, thread, agent and workspace.
  • A usage gauge for the active account in the status bar.
  • Context fill per agent, so you see which one is about to run out.
The Usage view showing three limits: Session (5 hours) 58% left, Weekly all models 32% left, and Weekly Opus 70% left, each with the time until it resets.

Workspaces

Folders, agents, capabilities and environments grouped together. Switch workspace and the terminals, files and git switch with you.

Agents

Each agent is a Claude Code session with a role and a model. Chat with it, approve its tools, queue messages, or stop a turn with Esc.

Crews that talk

Agents in a thread share a timeline and can message each other. Tag who a message is for with @, or choose a main agent that gets everything untagged.

Needs you

One place for everything blocked on a person: tool approvals, questions with numbered options, plans to confirm and messages addressed to you.

Terminals, git, files

Real terminals per workspace with scrollback. Stage, diff, commit, switch branches and push. Browse the workspace and open files in tabs.

Capabilities

Skills and MCP servers per workspace, with where each one comes from and roughly how many tokens it adds to context. Switch them on or off.

Usage and limits

Cost and tokens by model, thread, agent and workspace, plus your session and weekly limits and how long until each resets.

Accounts

Add more than one Claude login and switch the active one from the status bar. Sessions that are already running keep the account they started with.

More than one machine

Add another computer by its SSH host alias. Your daemon holds the tunnel, and threads on that machine show up in the same workbench.

Mods

Every feature is a mod, and yours load next to the built-in ones. Drop a folder with a mod.json into ~/.spawnkit/mods to add panels, commands or server code.

Themes

Light and dark built in. Import a VS Code theme and Spawnkit maps it onto its own palette, adjusting the accent so it stays readable.

Keyboard first

Command palette, go to file, new agent, new crew, terminal and source control are each one shortcut away.

03 For teams New

Share the setup. Keep the sessions local.

Set up a workspace once, with its agents, skills, MCP servers and repos, and everyone in your org can start a thread from it on their own computer. The org shares definitions. It never runs your sessions.

  1. Create an account

    You get a default org with it. Invite people later, or keep it to yourself.

  2. Set up shared workspaces

    Agents, skills, MCP servers and folders. Folders are GitHub repo links; each member clones them with their own git login.

  3. Invite your team

    Members join the org and its workspaces show up in their own workbench.

  4. Start a thread anywhere

    Anyone in the org starts a crew from the same setup on their own computer, with their own Claude login.

Your org on spawnkit.app

Shares and manages. Holds no sessions.

  • Shared workspacesAgents, capabilities and GitHub repo links
  • SkillsWritten once, used in the org’s workspaces
  • Hosted MCP serversRun on Spawnkit’s servers; their secrets stay there
  • Members and invitesWho can see and use the org’s workspaces
Your computer
  • Claude Code sessionsSigned in with your Claude login
  • ReposCloned with your own git login
A teammate’s computer
  • Same workspace, their sessionsTheir Claude login, their git login
Coming laterDedicated servers on paid plansSelf-hosted org hubFree while in beta.
04 Local first

Your machine does the work. Your data stays there.

Spawnkit is built to be boring about privacy. Here is exactly what stays local and what goes through spawnkit.app.

On your machine

Always, with or without an account.

  • The daemon listens on 127.0.0.1 and checks Origin and Host on every request and WebSocket.
  • Workspaces, transcripts and settings live in ~/.spawnkit, readable only by your user.
  • Spawnkit never reads your Claude credentials. The Claude CLI writes logins itself.
  • API keys and auth tokens inherited from your shell are removed from agent environments.
  • Nothing leaves your machine except what Claude Code itself sends.

Through spawnkit.app only with an org

If you create an account or join an org.

  • Your account and which orgs you belong to.
  • Shared workspace definitions: agents, skills, capability choices and repo links.
  • Calls to the org’s hosted MCP servers, which run on Spawnkit’s servers and are logged for the org.
  • Tool calls in the org’s workspaces, only if the org turns on its audit log. You are told before you add such a workspace.
05 FAQ

Questions, answered plainly.

Something missing? The getting started guide goes into more detail.

Do I need an account to use Spawnkit?

No. The daemon and the workbench run on their own, on your machine. An account is for the team side: an org, shared workspaces, hosted MCP servers and invites.

Does Spawnkit run Claude for me?

No. Every session runs on your machine through the Claude Code CLI, signed in with your own Claude login. Spawnkit starts those sessions, connects them and shows you what they are doing.

Usage counts against your Claude plan or account exactly as it would in a terminal.

What do I need to run it?

macOS or Linux on x64 or arm64, and the Claude Code CLI installed and logged in. Node is bundled with Spawnkit, so you do not need to install it separately.

Can my org see what my agents do?

Calls to MCP servers that your org hosts on Spawnkit go through Spawnkit’s servers, and those calls are logged for the org.

Orgs can also turn on an audit log of tool calls in their workspaces. When it is on, Spawnkit asks you to acknowledge it before you add the workspace, and the thread shows who it is recorded by. Outside of that, your code, terminals and transcripts stay on your machine.

Can I run agents on another computer?

Yes. Add the machine by its host alias from your ~/.ssh/config. Your local daemon opens and keeps an SSH tunnel to the daemon over there, and the machine picker switches the workbench between them. No VPN or extra network setup is needed.

Can I use more than one Claude account?

Yes. Add accounts in Settings, log each one in with the Claude CLI, and switch the active one from the status bar. New turns use the active account; running sessions keep theirs.

Two optional automations, switching accounts at a limit and a keep-alive ping, are off by default because they can conflict with Anthropic’s terms. You have to opt in to each.

What does it cost?

Spawnkit is free while it is in beta. Paid plans with dedicated servers will come later. Claude usage is billed by Anthropic under your own plan, as usual.

Can I extend it?

Yes. Every built-in feature is a mod, and your own mods load next to them from ~/.spawnkit/mods. A mod’s server code runs with the daemon’s full privileges, so install only mods you trust.

Start your first crew.

Create an account to get your org, or install the daemon and start on your own machine. Both are free while Spawnkit is in beta.

Already have an account? Sign in or open the app.