Type "use Claude for project management" into any search engine and you get the same advice three times over: open Claude Projects, upload your brief, ask it to draft a Work Breakdown Structure. Useful for an afternoon. Then Monday comes, the WBS is a stale PDF nobody opens, and you're back to updating the board by hand.
The direct answer: using Claude for project management works best when Claude has read/write access to your actual board, not just a chat window to draft documents in. Connected via MCP, Claude can triage cards, draft new ones from meeting notes, move stale work, and run on a schedule — not just write plans about the work, but touch it. Here's the real workflow, the parts that make it stick, and the honest limits.
What "using Claude for project management" usually means — and what it should
Most of what ranks for this query treats Claude as a planning layer: draft the charter, generate the timeline, turn a messy brief into a clean scaffold with objectives and deliverables. That's genuinely useful — Claude is good at turning fuzzy input into a structured first draft. But it's a one-way trip. You paste it into Claude, get text back, and then copy that text somewhere a human still has to maintain by hand.
What changes the equation is giving Claude the same access a new hire gets: a login to the actual tool, an account with its own identity, permission to move things. Not a document generator you consult once, but a board member you leave work with. That's the version this guide covers — how to set it up, what to hand it first, and how to make it a habit instead of a one-time trick.
How do you actually connect Claude to a real board?
The mechanism is MCP (Model Context Protocol) — an open standard that lets Claude call tools on a service directly, instead of you copying its output somewhere else. In Comuna, you add the connector once (Claude Settings → Connectors, OAuth, no API keys) and Claude appears in your workspace as a member with its own avatar and badge. We covered the exact steps in the 60-second setup guide — this post picks up from there: what to actually do with the connection once it exists.
Once connected, Claude has access to the same actions a teammate has: create cards, move them across columns, mark them done, write comments and notes, and read the board to answer questions grounded in your real data instead of a guess.
What should you delegate first?
Don't start with "manage my project." Start narrow and specific — the same way you'd onboard a new hire.
Triage the backlog. Point Claude at a messy column and ask it to tag, prioritize, or split anything that's really two tasks stapled together. This is low-risk (nothing gets deleted) and shows you fast whether its judgment matches yours.
Turn notes into cards. Paste — or let it read — a meeting transcript and ask for cards, one per action item, with an owner guess and a due-date guess. You edit, don't create from scratch.
Chase staleness. "Anything untouched in 10 days in In Progress — comment asking for a status, or move it back to Backlog if it's actually blocked." This is the task every PM guide assumes a human does manually. Claude can do the check every single day without getting bored of it.
Write resolution notes. When a card closes, a one-line note on what actually happened is the thing everyone skips under deadline pressure. Make it a standing instruction and it stops being optional.
None of this requires Claude to make a judgment call about strategy. That's deliberate — the riskier decisions come with an approval step, covered below.
How do you make it a habit instead of a one-off chat?
This is the part almost every "use Claude for PM" guide skips, and it's the difference between a novelty and a workflow: Claude Cowork scheduled tasks. On a paid Claude plan (Pro, Max, Team, or Enterprise), you can tell Claude to run a prompt on a recurring cadence — daily, weekday mornings, weekly — without you opening a chat window at all.
The pattern we use ourselves: a scheduled prompt each morning that says roughly "check the board, do the triage and staleness pass, draft cards from anything new in the notes, and give me a short summary." Claude runs it in the cloud on its own schedule — your laptop can be closed — reads the board through MCP, does what's safe, and stops at anything that needs your call. You wake up, check the daily brief waiting for you, and either everything's handled or three things are waiting for a decision.
That's a materially different habit than opening Claude Projects when you remember to. It runs whether you remember or not, and it only bothers you with the summary — not with 20 minutes of "let me check what happened."
How do you keep control of what it does?
Two mechanisms, and they're the part that makes handing over real board access defensible instead of reckless.
Escalation, not autonomy. When Claude hits something it isn't confident about — closing a card, reassigning someone else's work, anything with judgment attached — it doesn't guess. It opens a small approval request instead of acting. You approve, ask for changes, or reject. Over time it learns which calls you actually want to make yourself and escalates less on the ones you don't.
Signed attribution. Every card Claude touches, every comment it leaves, carries its name and badge — never folded silently into your account, never an anonymous "system" edit. If a card moved in a way you didn't expect, you can tell in one glance whether that was Claude or a teammate. Six months from now, when you're trying to reconstruct why a decision was made, that trail is the difference between an answer and a guess.
Be honest about the "pull, not push" part. Claude doesn't watch your board continuously and react in real time. It acts when you trigger it — from a chat, or from a scheduled task running on its own cadence. That's a real constraint, not a marketing footnote: if something urgent happens between scheduled runs, Claude won't jump on it until the next trigger. Set the cadence with that in mind.
Where this doesn't fit
If your team leans on a deep bench of native integrations — a specific Jira sync, a Salesforce workflow, years of Zapier chains — check what's actually connected today before assuming it carries over; younger MCP-native tools are narrower there than incumbents that have had a decade to build out an app marketplace. And if what you actually want is long-form document collaboration — specs, wikis, shared drafts as the primary artifact — a docs-first tool with project management bolted on may still fit your team better than a board-first one.
Scheduled tasks also depend on your Claude plan; confirm current availability on Anthropic's side, since plan details change independently of this post.
Is this the same as the "use Claude for project management" advice everywhere else?
No. Most of that advice stops at drafting a plan in a chat window — one-shot, disconnected from the tool you actually work in, with a human moving the text into the real board by hand afterward. Connecting Claude to the board over MCP means it reads and writes the same cards your team does, with its own identity, so the plan and the execution live in one place instead of two.
Do I need to be technical to set this up?
No. The connector is OAuth-based — click connect, sign in, authorize. No API keys, no config files, no code. The 60-second setup guide is the exact click path.
Can Claude run my board with no human involved?
No, and that's intentional. MCP is pull: Claude acts when triggered, either directly or on a schedule you set. Anything it's not confident about goes into an approval queue instead of happening silently. See what MCP actually changes for the fuller version of that distinction.
What if I also use ChatGPT?
Both can be connected to the same board at once, each with its own identity and badge, so their edits stay distinguishable from each other and from your team's. See the best PM tool for Claude comparison for how the two stack up.
Comuna is free forever — no credit card, bring your own AI. Spin up a workspace and try it.