ChatGPT for project management works in two different ways, and most guides only describe the first. In the common one you paste your tasks into a chat window, get a plan or a status summary back, and retype the useful parts into your project tool. In the second, ChatGPT connects to your board through a protocol called MCP, reads the cards that are on it right now, and creates or moves cards when you ask. This post covers both, and where the first one stops being worth the effort.
What most ChatGPT project management guides teach
The standard advice is prompting, and it does work. ChatGPT is good at turning a rough brief into a work breakdown, drafting a risk list, writing the status update nobody wants to write, or converting a page of meeting notes into something that looks like tasks. Entire courses are built on this. The Coursera specialization on ChatGPT for project management covers planning, execution and prompt patterns without ever connecting ChatGPT to a project tool.
A typical prompt from that school of advice looks like this:
Here is my project brief. Break it into phases with tasks, dependencies and rough estimates. Output a table I can paste into my tracker.
You get a decent table. Then you paste it into your tracker by hand, and from that moment the table and the tracker start drifting apart.
Where copy and paste runs out
The chat window has no idea what happened after you closed it. Every new conversation starts from nothing, so you re-explain the project, re-paste the current status, and hope you pasted the current version. Two weeks in you are maintaining the board and a private summary of the board that lives in your chat history.
That is the real cost of prompt-only project management. Not the writing, which is fast, but the bookkeeping around it. Ask "what should I do today" in a fresh chat and the answer describes a project that existed the last time you pasted it.
What changes when ChatGPT can read the board
MCP is the piece that removes the pasting. The Model Context Protocol is an open standard for connecting AI applications to outside systems, and OpenAI supports it in ChatGPT through custom connectors. A tool exposes an MCP server, you connect it once, and the model can request live data when a question needs it.
For a kanban board that means ChatGPT can list your boards, read the cards in each column, and take actions you approve. The board stays the source of truth. Nothing is copied anywhere, and the answer you get on a Thursday reflects the card someone moved on Wednesday.
How to connect ChatGPT to a kanban board
EasyKanban runs a hosted MCP server, so there is no server to install and no API key to paste. The connection uses your existing login through OAuth, and it inherits the permissions you already have. If you are a viewer on a shared board, the connector is a viewer too.
- In ChatGPT's settings, open the connectors section and choose to add a custom connector. OpenAI moves this menu around, so their MCP documentation is the place to check the current path.
- Paste the server URL and confirm the EasyKanban sign-in when the browser window opens.
https://easykanb.app/api/mcp
The /api/mcp part matters. The website address on its own is not an endpoint. Client-specific instructions for Cursor, VS Code and the desktop apps live on the MCP setup page.
Then test it with a question that changes nothing:
List my EasyKanban boards and tell me how many open cards are in each column. Do not change anything.
If the counts match what you see in the app, the connection is working and pointed at the right account.
What to ask once it is connected
The useful prompts are the ones that would otherwise cost you fifteen minutes of reading columns. Three that earn their keep:
Morning triage. "Look at my Client Work board. Which cards in Backlog have been sitting there longest without a due date? List five, oldest first, and do not move anything." You get a shortlist instead of a wall. Weekly review. "Summarize what moved to Done on all my boards in the last seven days, and name the cards that have been In Progress for more than five days." This is the review most people skip because collecting the information is the boring part. Notes into cards, with a checkpoint. "Turn these notes into cards for the Product board. Show me the titles and target columns first and wait for my confirmation." The two-step pattern is worth the extra message. It lets you fix a bad title before it exists rather than after.Vague instructions are where this goes wrong. "Clean up my boards" could mean archiving stale cards, merging duplicates or deleting things, and you will not enjoy finding out which one it picked. Name the action you want.
What it still gets wrong
The model only knows what the card says. A card called "Improve onboarding" gives it nothing to work with, while "Add a password-reset hint below the login form and check it on mobile" gives it a target and a finish line. Better cards make the AI more useful, and they make the board more useful to you as well, which is the honest argument for writing them.
Prioritisation stays yours. ChatGPT can tell you a card has sat in Review for nine days. It cannot tell you that the client stopped answering emails, unless somebody put that on the card.
There is also a plan boundary worth knowing before you start. Reading boards and to-do summaries are free and unmetered on EasyKanban. Writes are free up to 50 per calendar month, and unmetered on Pro. The free plan holds three boards with unlimited cards, so a solo backlog fits comfortably inside it.
ChatGPT or Claude for this?
Both speak MCP, and the EasyKanban server is the same endpoint for both, so use whichever assistant you already pay for. The setup steps differ only in where the connector settings live. The Claude walkthrough is written up separately in connect Claude to your kanban board, and the prompt library in AI prompts for project management applies to either client.
If you are still deciding whether AI belongs anywhere near your task list, AI task management with kanban is the slower introduction to the same idea.
Frequently asked questions
Can ChatGPT actually update my kanban board?
Yes, once you connect a board through an MCP connector. ChatGPT can create cards, edit them, move them between columns and archive them. Each write happens through a tool call your client asks you to approve, and your existing board permissions still apply, so a read-only member stays read-only.
Is it safe to connect ChatGPT to my project data?
The connection uses OAuth against your existing login, so no password or API key is shared with the AI client. Access matches the permissions your account already has, and you can revoke a connected app at any time from Settings without changing your password or affecting other clients.
Do I need a paid plan to use ChatGPT for project management this way?
On the EasyKanban side, reading boards and to-do overviews is free and unmetered. Writing is free for 50 actions per calendar month and unmetered on Pro, which costs 6 euros a month. Whether your own ChatGPT subscription can add custom connectors is a question for OpenAI's current documentation.
What can ChatGPT do that a normal kanban board cannot?
It reads across every board at once and answers questions a column layout cannot show, such as what has been stuck longest, or which cards nobody owns. The board is still where the work lives. The connector removes the manual reading you would otherwise do to answer those questions.
Will ChatGPT delete my cards by mistake?
Destructive actions are marked as such in the protocol, so clients prompt before running them. The practical protection is still your own wording. Ask for a list of proposed changes before any write, and never hand over an instruction as broad as "tidy this up".
Start with one board and one question
Connect a single board, ask for a read-only summary, and compare it against what you see in the app. That comparison tells you whether your cards carry enough context to be worth asking about, which decides how useful the rest of this will be. Add write actions later, one at a time, once the answers stop surprising you.
The board keeps the work. ChatGPT just saves you from reading all of it every morning.