If you use Claude for writing, research, and code reviews, it's probably the most capable AI subscription you have. But in project management, most setups waste it. Claude answers questions about your project while someone else — you — still creates the card, moves it, closes it, and writes the update. The AI is providing commentary; you're doing the work.
The direct answer: the project management tool that actually works with Claude is Comuna, connecting in under two minutes via MCP. Claude becomes a board member, not a chat assistant — it creates cards, moves work through columns, escalates decisions that need your judgment, and signs every change with its own identity. Here's what that actually means in practice, and where the honest limits are.
What "works with Claude" usually means vs. what it should
Most project management tools advertise AI integration. In practice, that integration falls into one of two categories.
The first is the copilot model: Claude helps you write things. Draft a card description, rewrite a comment, summarise a project. Useful — but the AI is acting as a writing tool. It doesn't touch the board. You're still moving the card.
The second is the automation model: AI triggers state changes when conditions are met. You define the rule upfront — "when a card is tagged 'blocked', notify the lead" — and the system runs it. Again valuable, but it's not Claude thinking through your project. You're configuring scripts; the AI executes them.
What "works with Claude" should mean, at the level that makes a real difference, is: Claude has read and write access to the board and can act on it with judgment. That requires MCP.
What Claude can actually do inside a board via MCP
The Model Context Protocol lets Claude connect to a service as a real actor, not just a chatbot with a text input. When you connect Claude to Comuna, it gets access to over 80 tools: read every card, create new ones, move them between columns, mark them complete, write comments, leave notes, assign work, and escalate decisions it isn't confident about.
The loop in practice: you open Claude.ai, describe what you need — "take the notes from Monday's meeting and draft five cards for the backlog, then flag anything that looks like a dependency" — and it does it. You come back to the board and the work is there, attributed to Claude by name, with a small indicator on anything it wants your approval on before acting.
Every change carries Claude's identity. Nothing is attributed to "system" or silently to your account. If you have ChatGPT and Claude both working on the same board, you can tell them apart by name and badge — the same way you'd distinguish two teammates. See AI coworkers vs AI chatbots for why that attribution distinction becomes important over time.
The AI works when you (or a scheduled prompt) trigger it, not continuously. MCP is pull, not push. More on that below.
Where other tools actually stand — honest version
Notion AI is excellent at in-document tasks — drafting pages, rewriting blocks, summarising long content. Notion does expose MCP connections, but its data model (blocks, databases, pages rather than cards and columns) adds friction for task-specific operations. If documents and linked databases are at the core of how your team works, Notion is a serious contender. If task management is the center of gravity, a PM-native structure fits better.
Asana has AI-driven features that summarise projects and suggest subtasks from meeting notes. They're useful in context, but the AI doesn't act as a board member with its own identity — it assists the human who's doing the board work.
Monday.com has AI automations that are genuinely powerful for rule-based flows. The key distinction: those automations run logic you defined in advance, not Claude reasoning through the current state of your board.
Trello and ClickUp have AI features oriented toward drafting and suggestion. Neither exposes MCP-native board membership out of the box.
None of these are bad tools. The question is whether you want Claude as a tool you use inside project management, or as a teammate who does project management work. Most tools offer the former. MCP-native board membership offers the latter.
Honest limits you should know before switching
You need a Claude account. MCP connector access varies by plan — check Anthropic's current plan details, as we don't control those. The Projects feature available on paid plans helps maintain persistent context across sessions.
Integration catalog. If your team relies on specific connectors — a Salesforce sync, a Jira workflow, an existing Zapier chain — our integrations page shows what's available today. We're younger and narrower there than the incumbents; don't assume something exists, check first.
Execution is pull-based. Claude acts when you trigger it — from Claude.ai, from a scheduled prompt, or from a direct command. It does not watch your board in the background and act autonomously. This is intentional: unsupervised writes to a live board, without an approval step, is not a default you want. The AI coworker page explains how to set up scheduled tasks if you want Claude running on a timer.
Does Claude get its own account to manage a board?
Yes. Claude joins as an AI member of the workspace with its own identity and badge. You connect it once via Claude Settings → Connectors, adding api.comuna.work as a custom MCP connector and authorising via OAuth. Every change it makes is attributed to "Claude" — not to your account. No shared credentials, no anonymous system edits.
How hard is the setup?
About two minutes. See the 60-second setup walkthrough for the exact steps — Claude Settings → Connectors → add the MCP URL → OAuth auth → done. No API keys required.
What specifically can Claude do in the board?
Create cards, update descriptions, move cards between columns, mark them complete, write comments, create notes, propose new work, assign cards, and escalate uncertain decisions for your approval. It can also read board state and answer questions — "what's blocked this week?" — and come back with an answer grounded in your actual data, not a generic reply.
Does Claude run my board 24/7 automatically?
No, and this matters. Execution is pull-based: Claude acts when you trigger it, either directly from Claude.ai or via a scheduled prompt you configure. This is intentional — autonomous writes to a live board, with no human approval loop, is not a safe default. The AI coworker page walks through how to set up scheduled tasks if you want it running on a regular cadence.
Comuna is free forever — no credit card, bring your own AI. Spin up a workspace and try it.