Claude Code Project Management: Give It a Live Plan
A coding agent is only as good as the task it is given. Connected to a live plan, Claude Code can pick up the next task, read its acceptance criteria, report progress and leave an audit trail, without anyone copying and pasting context.
Claude Code can work from a live project plan when connected over MCP. It lists its assigned tasks, reads acceptance criteria and dependencies, updates status, logs activity and writes a final summary, using a scoped API key and a generated work order so scope stays controlled.
- Plan in, status out: the agent reads tasks and acceptance criteria from the project and writes status and activity back.
- Work orders for precision: a generated brief covers identity, project context, story, acceptance criteria, operating rules and API reference.
- Scoped and auditable: a per-agent API key limits what it can touch, and activity logs show what it did.
- Humans stay accountable: scope changes go through change requests; the PM reviews the final summary.
The Context Gap Between a Plan and a Coding Agent
Most coding agents start every session blank. You paste a ticket, a few notes and hope. The agent then cannot see the dependencies, the acceptance criteria or what else is in flight, and you cannot see what it did except in a diff.
Connect it to the plan and both sides of the gap close: the agent sees what you see, and you see what it reports.
Connect Claude Code in Two Commands
- Create a scoped API key (see MCP Server for Project Management).
- Run
claude mcp add --transport http comng-ai https://comng.ai/app/api/external/mcp --header "X-API-Key: <key>".
The Work Order: A Brief That Is Actually Complete
Register Claude Code as an AI agent on the project team, then use AI Offload / Work Orders to turn a task, or a whole branch of the work breakdown, into a fully specified brief. The work order includes:
- Identity: who the agent is on this project.
- Project context and goals: what the work is for.
- Story overview and task details, including the acceptance criteria.
- Connection instructions: MCP recommended, REST as a fallback.
- Operating instructions: what to do at session start, where output goes, during work, and the mandatory update after every task, ending with a final summary note.
The Loop: Pick Up, Work, Report
Pick up
The agent lists tasks assigned to it and reads the dependencies and acceptance criteria.
Work
It builds in your repository and uses the project tools for context as needed.
Report
It updates task status, logs an activity with time and notes, and appends to a note.
Summarise
At session end it writes a final summary note for the project manager to review.
What the Project Manager Sees
Because the agent is a team member, its work appears everywhere human work does: the task list, the Kanban board, the Resource Planner and Project Pulse. An agent loaded at 360% shows as a red cell exactly as a person would. See The 24-Week Burnout Radar. If the agent finds that the task is out of scope or blocked, the right path is a blocked flag with a reason or a change request, not a silent workaround.
| PM question | Where the answer is |
|---|---|
| What did the agent do? | Activity log and the final summary note |
| Is it stuck? | Blocked flag with reason; Pulse stalled-task advisories |
| Is it overloaded? | Resource Planner heatmap |
| Did it touch anything it should not have? | Key permissions limit it; Audit Logs record actions |
- Paste a ticket and hope
- No record of what the agent did
- Agent invisible in capacity planning
- Agent reads the live task and criteria
- Status, activity and summary written back
- Agent workload on the same heatmap as humans
About CoMng.AI
CoMng.AI is an AI-native project management platform that generates a project's plan, tasks, risks and budget from a description or document, monitors project health in real time, and gives both human teams and AI coding agents live access to the same project data, with a person approving every AI-generated change.
Give your AI agents a live plan and a safe way to work in it
Create a scoped API key, connect your agent over MCP and let it pick up work from the same project your team uses.
Frequently Asked Questions
Can Claude Code connect to a project management tool?
Yes. Through the CoMng.AI MCP server and a scoped API key, Claude Code can list tasks, read acceptance criteria and dependencies, update status and log activity.
What is a work order?
A generated brief for a task or a branch of the work breakdown containing the agent's identity, project context, story and task details with acceptance criteria, connection instructions and operating rules.
How do I limit what Claude Code can change?
Create a dedicated API key with only the actions it needs, restricted to the relevant project, with a rate limit, expiry and an IP whitelist.
Does the agent show up in capacity planning?
Yes. AI agents are team members, so their workload appears on the Resource Planner and in Pulse alongside people.
Who approves scope changes the agent finds?
A person. Scope changes go through change requests, which require approval before they are applied.
Related reading
See the full guide: AI Agents, ARI & MCP Guide