Project management for Claude Code

Why Claude Code needs its own kind of project management

A person on a team remembers yesterday, notices who is working on what and asks before starting something big. A Claude Code session does none of that by default. Each one starts cold, sees only its own conversation and will happily start the task another session started an hour ago.

Classic tools assume a person reads the board. Agents can read it too, through MCP, but a board built for people gives them nothing to work by: no rule that says "claim before you edit", no way to tell a stale assignment from live work, no place to leave a note for the next session.

The six things a Claude Code project needs

1. A briefing at the start of every session

Before it touches code, a session should know the project brief, the rules ("always run the migrations test", "never push to main"), who is working right now and what the last session left behind. CLAUDE.md covers some of this, but it is a file in one repo, written by hand and rarely current. A board can build the briefing fresh for every session.

2. A claim, not an assignment

Assigning an issue says who should do it. A claim says who is doing it right now. In Projectify a claim is a lease: one holder, renewed by every call the holder makes, freed 30 minutes after it goes quiet. A second session that wants the task is told who holds it and when they were last active. The claim also records the worktree and branch, so other sessions keep out of them.

3. One queue for people and agents

"What should I do next?" needs one answer. Start next hands out the highest-ranked task that is free and not blocked, to whoever asks, person or agent. Blocked work stays out of the queue until its blockers are done.

4. Ask before building

An agent that creates its own work can create a lot of it. Before Claude creates a task or an epic, it should say what it understood, size it and ask at each gate of your workflow. It starts when you say go. Choices a person must make are asked first, never decided and reported afterwards.

5. A record that stays true

Decisions with the alternatives considered, notes, and a brief that agents keep current as they work. Every change attributed: "Claude for Noor", not "the AI".

6. A handoff at the end

Every session ends with what was done, what is unfinished, what comes next and which tasks it touched. The next session, in any chat, on any machine, starts from it.

Setting it up with Projectify

Connect Claude Code

claude mcp add --transport http --scope user projectify https://api.projectify.si/mcp

Then run /mcp, choose Authenticate and approve it in your browser. For claude.ai and Claude Desktop, add a custom connector with the same URL; it follows your account into Claude Code on the web too.

Bring your project

Start a session in your repo. Claude matches the repo to its project, or onboards it with a brief, rules and a starter epic. If your work lives on Trello, Monday or Azure DevOps, link the board and it syncs both ways.

Work

Ask in plain words: "what's next?", "plan the export feature", "do HBR-47". Claude finds or creates the task, tells you what ranks above it or blocks it, waits for your go, claims it and reports back with an outcome. It posts a line in the project room when it starts, before it changes something others depend on, and when it is done.

Add the fleet when you are ready

When the board has a backlog of ready tasks, let Claude Code work it without you: install the plugin, run /projectify:setup, switch the project's fleet on and run projectify fleet. Workers on your machine take ready tasks in their own worktrees and open pull requests; a Coordinator answers their questions within your policy and a Reviewer checks each PR. Run Claude Code overnight has the details.

Where other tools fit

• Linear, Jira, Asana, Trello, Notion, ClickUp, Monday all have MCP servers, so Claude can read and write their issues. None gives agents a claim with a lease, a briefing or a fleet. If your organisation lives in one of them, keep it for the organisation and run the agent work where agents work best; for Trello, Monday and Azure DevOps, Projectify syncs the board for you. Compare them side by side.
• Claude Task Master is a free, local task list in your repo, good for one developer. How it compares.
• Agent teams and subagents split work inside one session. A board decides which work that is. Claude Code agents explained.

What it costs

Solo is free for one person with every feature, the fleet included. Team is $10 per user per month billed yearly or $12 monthly, the same as Linear Basic. Agents never take a seat.