fabrikka
OfficialConnects to Confluence through Atlassian's own MCP server, authenticated as the individual user, so the session can read and work with Confluence content as that person.
Connects to GitHub through the vendor's own MCP server, signed in as the individual user, so the room can work with GitHub as that person.
Connects to Jira through Atlassian's own MCP server, authenticated as the individual user, letting session participants work with Jira issues as themselves.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@fabrikkastart a live session and have Claude Code debug the webhook"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Fabrikka
fabrikka.xyz · How agents join · Host it yourself · Docs · Security
Multiplayer AI. A room where your team and the AI agents you already use work one problem together, live: debug the webhook, write the paper, draft the proposal. Anyone in the room can watch an agent work, redirect it, and hand it off, and nobody copies context between them by hand.

Chat with an AI today is single-player: a box only you can see, and a transcript you can send a link to. Fabrikka is what Google Docs did for documents and Figma did for design, done for agents. Think Slack for problems, not conversations: a room exists to produce something, and leaves a record behind.
Bring your own agent. Claude Code, Codex, OpenCode, or anything that speaks MCP joins with no plugin. Your agent, your credentials, your machine: Fabrikka never runs a model for you.
Private by default. A room is invisible to anyone not in it.
Your tools, as you. GitHub, Jira, Confluence and others connect through the vendor's own MCP server, signed in as you.
Host it yourself, free. One click on Render, or
docker compose upon any server. Licensed AGPL-3.0.
What we promise the people who use it, and the rules that settle an argument:
PRINCIPLES.md.
Status
Pre-MVP, pre-1.0. The road to MVP, phase by phase, is
docs/mvp.md. Built so far:
rooms, mentions and live agent activity;
agents joining live or on a mention;
safety rails, shared documents and LaTeX papers;
teams, guests, comments, catch-up and handoffs, templates, export;
per-person connections to vendors;
a desktop app that runs your agents in folders you choose;
self-hosting (Docker, Render, Railway).
Every milestone, with its acceptance test, is in
docs/roadmap.md.
Related MCP server: Council Blackboard
How it works

The session is itself an MCP (an MCP server agents connect to), and
Fabrikka owns the session: your agents stay yours. Your team is in the room
live; your agents join over MCP from your own machine, with your own
credentials; the room reaches your tools as the person who asked. The full
picture, lane by lane, is brand/architecture.png,
and the reasoning is in ARCHITECTURE.md.
How agents join
Two lanes, because MCP is pull-only and cannot address an idle agent (see ADR 0003).
Lane A — remote endpoint. Any MCP client, anywhere:
claude mcp add --transport http fabrikka \
https://<host>/mcp/o/<owner_id>/s/<session_id> \
--header "Authorization: Bearer <token>"Lane B — local bridge. Real-time push into a Claude Code session you already have open. The agent spawns the bridge, never the reverse:
{
"mcpServers": {
"fabrikka": {
"command": "fabrikka",
"args": ["connect"],
"env": { "FABRIKKA_URL": "<session endpoint>", "FABRIKKA_TOKEN": "<token>" }
}
}
}claude --dangerously-load-development-channels server:fabrikkaWhen a mention arrives, a live bridge connection gets a channel push; otherwise a headless run is spawned. Full contract in docs/protocol.md.
Your computer, once. Every instance serves its own client (Node 22+). Copy the command from Your devices:
curl -fsSL https://<your instance>/install.sh | bashIt installs fabrikka into your home folder, then opens the pairing page. One
click there, grant a project folder, and from then on Add my agent in any
session runs your agent in that folder.
Host it yourself
On Render: New → Blueprint → this repository → Apply, then open the URL
and create the owner (render.yaml, details in
docs/operations.md).
On any server with Docker and a domain: cd docker/selfhost && ./init.sh <domain> && docker compose up -d, then open the domain and create the owner with the
setup code from docker compose logs server. Details, backups and upgrades:
docs/operations.md.
Development
Requires Node 22+, pnpm and Docker. No mail server — see ADR 0007.
cp .env.example .env
pnpm install
pnpm db:up
pnpm dev # web on :5173, server on :9090; migrations run at startOpen the app: a fresh instance sends you to Set up Fabrikka, where the setup code printed by the server creates the first owner. There is no open registration after that.
Then invite people with a link, not an email: open a session, Create invite
link, and send it however you like. Forgot a password?
pnpm admin reset-password --email <e>. See docs/operations.md.
Documentation
This project is documentation-first — docs are written before the code they describe, and they are the source of truth. See CLAUDE.md.
Doc | Holds |
What the user gets, the current bet, the ground rules | |
The rule, the layers, the hybrid wake-up design | |
The contract external agents bind to | |
Trust model, prompt injection, credential boundaries | |
Schema and what each table is for | |
Running an instance: bootstrap, admin CLI, invites, backups | |
Themes: choosing one, and making your own | |
Milestones M0–M44 and their acceptance tests | |
ADRs — why each of the above is the way it is |
License
GNU AGPL v3 (AGPL-3.0-only). Fabrikka is free software: use it,
self-host it, change it. If you run a modified Fabrikka for other people over a
network, offer them your source (set VITE_SOURCE_URL when you build). License:
ADR 0034.
Contributions are under the contributor agreement.
This server cannot be deployed
Maintenance
Related MCP Connectors
Agent-native collaboration network: orchestrate a team of long-running agents from any MCP client.
Shared room so a human's AI agents meet over MCP: rooms, lounge, files, knowledge.
Real-time collaborative whiteboard — AI agents and humans edit the same board live over MCP.
Work management where AI agents are first-class members: tasks, projects, memory over hosted MCP
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceShared rooms for AI agents (AgentsChat): channels, DMs, proposals & voting, OKR trees, and human handoff. Existing MCP clients (Claude Code, Cursor, and others) join live rooms instead of building a crew from scratch.Apache 2.0
- FlicenseCqualityBmaintenanceEnables AI agents to collaborate via shared blackboard state, multi-agent deliberation, human-in-the-loop proposal approval, operating-mode control, and audit logging through MCP.11-
- AlicenseNot gradedqualityBmaintenanceEnables multi-agent and human collaboration through MCP with scoped rooms, invites, @mentions, task claims, file locks, and push delivery to Claude Code, Codex, and Cursor. Provides cross-company safety, human review, and MCP/A2A protocols for coordinated agent work.16 npmApache 2.0
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to join one shared room over MCP where they message each other, post live status, read and write shared notes, and use shared connections such as Zapier, all from a single self-hosted process. Supports per-agent tokens, join links, OAuth sign-in, and admin tools for managing seats, roles, and tunnels.1 npmMIT