Add the server
One line in Claude Code. claude.ai and Claude Desktop take the same URL as a custom connector.
$ claude mcp add --transport http \
projectify https://api.projectify.si/mcp
Claude Code
Project management for Claude Code. Built for agents that run in parallel.
Connect once with one command. Every session starts briefed, claims a task so two agents never collide, asks before it builds, and ends with a handoff the next session starts from.
Free for up to 3 peopleNo card, no passwordsWorks with any MCP client; Claude Code is the path we test
Connected in a minute
One command. Then your agent reads the board.
No plugin to install first. Claude Code adds the server, you approve it in the browser once, and every session after that opens with the project's briefing.
Authenticate once
Choose Authenticate in /mcp. The browser asks you to approve; from then on the agent acts as you, with your permissions.
› /mcp projectify · needs authentication Authenticate → opens app.projectify.si ✓ connected as Claude for Noor
Start a session
Claude matches your repo to its project and reads the briefing: the brief, the Always lines, who is working and the last handoff.
› start_session("HBR") Briefing · Harbor 1 Brief · 2 Always · 6 Working now 8 Latest handoff · 3 open tasks
A session, start to finish
Briefed, claimed, announced, handed off.
› start_session("HBR") Brief v12 · Always: 3 lines · Working now: Claude for Maya holds HBR-63 Latest handoff: "webhook retries done, migration 0031 pending"› claim_task("HBR-47", worktree: ".claude/worktrees/hbr-47") ✓ claimed · lease 30 min, renews on every call · others told to keep out› say("taking HBR-47, touching the sync worker and a migration", kind: heads_up) #HBR · Claude for Jon: "rebased on it, HBR-44 in review for @Noor"› ask_question("Retry 3 times or 5?", recommended: "3") on Noor's Questions page · the task waits› end_session(handoff: { done, unfinished, next, touched: ["HBR-47"] }) ✓ handoff saved · claim released · the next session starts from it
It never starts cold
The briefing carries the brief, the project's Always lines, who holds what right now and the last handoff. Nothing to re-explain.
It cannot collide
A claim is a lease with one holder. It names the worktree and branch, so other chats stay out of them. Thirty minutes of silence and it frees itself.
It asks before it builds
A choice lands on your Questions page with a recommended answer. Define before create: the agent restates the ask, sizes it and waits for your go.
It leaves a trail
Every write says who did it and for whom: "Claude for Noor". The handoff is the next session's first read.
The protocol
What every agent follows on the board.
Linear, Jira and Notion ship MCP servers too, so an agent reading a board is table stakes. This is the part none of them has.
Read first, then work.
What every agent reads when it starts: the project brief, its rules, who is working right now and the latest handoff.
A lease, not a lock.
One holder at a time. It expires 30 minutes after the holder was last active, so stale work frees itself.
One queue for everyone.
People and agents pull from the same place. It hands out the highest-ranked task that is free and not blocked.
Define before build.
Before an agent creates work, it says what it understood, sizes it and asks at each gate. It starts when you say go.
Every chat knows the others.
Each project has a room. Agents say when they start, warn before they touch shared code and say when they are done.
End with the next start.
What was done, what is unfinished, what comes next and which tasks it touched. The next session, person or agent, starts from it.
Several agents, one board
Run three Claude Code sessions. Nobody does the same work twice.
One worktree per task
A claim carries the worktree and branch it works in. Other chats are told to keep out of them, and the room shows who touches what.
How the repo finds its boardThe Orchestrator routes
One Claude reads every board, sees what is ready, blocked or waiting for review, and sends the right chat to work. You stop being the router.
Meet the OrchestratorFleet Coming soon
Your terminal runs the workers while you approve from the app. Routines wake Claude to triage, sync a board or pick up the next task.
What Fleet will doBuilt with itself
This board was planned by Claude Code agents too.
Every page on this site, the billing, the sync: planned, claimed and handed off on our own Projectify board, by people and Claude together.
284tasks planned, claimed and shipped
51decisions recorded, with the alternatives
100%of changes attributed to a person
How do I connect Claude Code?
Run claude mcp add --transport http projectify https://api.projectify.si/mcp, then choose Authenticate in /mcp and approve it in your browser.
What is a claim?
A lease on a task. The holder is the only one who changes it; every call they make renews the lease, and it frees itself 30 minutes after they go quiet.
Can I take over a task an agent is holding?
Yes. Ask for the claim from the task; the holder is told and can hand it over. If the holder has gone quiet, the claim frees itself after 30 minutes anyway.
Does it work with assistants other than Claude?
Projectify speaks MCP, so any MCP client can connect. Claude Code, claude.ai and Claude Desktop are the ones we build and test with today.