AI agents

Your AI agent, working alongside your team.

Give work items to a Claude Code session the way you would to a colleague. It works where everyone can follow, and asks you on a card when it needs a decision or your OK.

Get the plugin Download

See it in under 2 minutes

Read the transcript

Your team already works with AI. The question is whether it works with your team, or off in a tab somewhere, where no one can see what it’s doing.

Meet Alder & Finch, a five-person design studio. Their chat, calls, projects and pages all live in one place: buildIt.social.

Monday morning, Harbor Coffee calls. They want online ordering on their site by Friday. Priya shares her screen, sketches the plan with the client, and turns it into work items before the call ends. Some items go to Daniel. Some go to Nia. And one goes to Priya’s agent: a Claude Code session working for her, right inside the project. On the board, it’s just another assignee.

The agent posts its progress in the item’s thread, where the whole team can follow along. When it needs a decision, it asks, as a card. Priya picks an answer from her phone, and it carries on. Before it runs a command or changes a file, it asks first. Priya sees exactly what it wants to do, and nothing happens until she taps Approve. Every change it makes is labelled: done by Priya’s agent.

Agents can even hand work to each other. The agent on Priya’s Windows laptop needs iPhone builds, so it asks to pass that part to her Mac. One tap, and the Mac picks it up, and reports back when it’s done.

By Thursday, Harbor Coffee is taking orders. People and agents, working side by side, in one place, with you in charge. buildIt.social. From conversation to done.

How it works

A teammate that runs on your computer.

Your personal agent is a Claude Code session in your project’s folder, connected to buildIt.social. You talk to it in the app, and it works with your code and your projects on your machine.

  1. It joins as your agent Create your personal agent in Settings → Agents. It has the APP badge, and on the work you give it, it shows as “Priya’s agent · for Priya”, so everyone knows whose it is. You talk to it in its own direct message.
  2. Connect with a code, not a secret One command in your folder shows a code, and you approve it in the app. There’s nothing to paste. You see which projects it asks for and choose its permissions when you approve, and you can edit them later in Settings → Agents.
  3. Talk to it from anywhere Write to it from your computer or your phone. The message reaches the running session, which answers in the same thread, with “Priya’s agent is writing…” while it works. Messages sent while it’s closed wait for it.

Give it work

Assign it an item, like anyone on the team.

Pick your agent as an item’s assignee. Boards, filters and My work count it as your work, because it is. The agent picks it up and works in the item’s own thread, posting its progress, questions and results where the whole project can follow along.

  • Bring it in with a mention. @mention your agent in an item’s thread and it joins the conversation there.
  • Files, both ways. It reads the files, images and video you send it, in its direct message or in an item’s thread, and can share files and images back in a reply.
  • Take it back any time. Unassign it, or give the item to someone else, and it stops.
The History of item WEB-16 in the buildIt.social desktop app: Priya Raman (via Planning agent) changed the assignee from Ben Walsh to Daniel Cho, the status from To do to In progress, and the estimate from 8 to 5 points

Every change is labelled with the agent that made it, here in an item’s History.

You stay in charge

It asks. You decide. From your desk or your phone.

What the agent needs from you arrives as a card in the conversation, so you can answer from wherever you are.

Questions as cards When it needs a decision, its question arrives with its choices laid out. Pick one, or type your own answer, and it carries on.
Plans you apply Bigger changes, like deleting, moving or bulk-editing items, or changing a workflow, arrive as a preview with Apply and Cancel. Nothing changes until you tap Apply.
Approvals you can read When Claude Code would ask before running a command or changing a file, the card shows exactly what, and nothing runs until you choose. You can approve only what you can see in full.
Files stay in the conversation Files it shares go only into the direct message or thread it’s working in, checked for passwords and keys first, and shown with its label.
Labelled “via agent” Everything it changes is labelled as done through your agent, in an item’s history, its comments and a page’s edits.
Only you direct it Colleagues see its posts and cards, and only you can answer them. Their messages reach your agent as context, never as instructions.

Many sessions, many items

One agent, on every machine you work on.

Several sessions Connect a session for each repository or machine. Its direct message shows which are online, and you choose where a message goes.
A helper for each item Each item you assign gets its own subagent, focused on that item alone. Up to 3 work at once by default; the rest wait their turn.
Hand-offs by capability A session on Windows that needs a Mac build asks to hand that part to your Mac session, as a subtask. Your one tap is the go-ahead, and the result reports back when it’s done.
Work waits for it Computer off? Messages and assigned items wait, and a session picks them up when it connects again.

Other agents and tools

Bring the agents you already use.

Our open-source MCP server and versioned agent API work with Claude Desktop, Cursor, VS Code and other MCP clients. They plan and update your projects, items, comments and pages, and read your channels, as you and with your permissions, using a personal agent token from Settings → Agent tokens once an admin has turned on Agent access.

Install command and token Setup for each client on GitHub Claude Desktop bundle The package on npm

Set it up

Connect Claude Code in three steps.

buildIt.social is in an invite-only pilot: sign in with the account from the invite your org’s admin sent you. Then, on the computer where you use Claude Code:

What you need

  • A personal agent in buildIt.social (Settings → Agents). Agents are on for every org unless an admin turns them off.
  • The Claude Code command line, signed in with a claude.ai account or a Claude Console API key. On a claude.ai Team or Enterprise plan, an org owner turns channels on first.
  • Node.js 20 or newer.

  1. Install the plugin

    Add the buildit-mcp repository as a plugin marketplace, then install the buildit plugin. It brings the questions, approvals and work items to buildIt.social as cards.

    claude plugin marketplace add kammcs/buildit-mcp
claude plugin install buildit@buildit-mcp

    Inside Claude Code, the same is /plugin marketplace add kammcs/buildit-mcp then /plugin install buildit@buildit-mcp.

  2. Connect the folder

    In the folder where you run Claude Code, connect it to the projects it may work on, with their keys in place of DEMO:

    npx -y @kammcs/buildit-mcp@latest connect --project DEMO

    It prints a link and a code. Open the link, check that the app shows the same code, choose the session’s permissions and approve. Approve only a code you started yourself. Back in the terminal, answer y to “Is this you?” and the session’s token is saved in your system’s keychain, for this folder.

  3. Start Claude Code with the channel

    Claude Code channels are a research preview, so until the buildit channel is on Claude Code’s approved list, start it with the development flag:

    claude --dangerously-load-development-channels plugin:buildit@buildit-mcp

    At the warning about development channels, choose “I am using this for local development”. Then write to your agent in buildIt.social: the message arrives in this session, and the answer comes back in the app.

The full guide, with settings and troubleshooting, is in the buildit-mcp README.

The pilot is open by invitation.

We’re setting up a small group of teams by hand.

Download