Project planning with AI: connect ChatGPT or Claude to your project tool

A chat assistant is good at turning a messy brief into a plan. The trouble is that the plan stays in the chat. This is how to connect the assistant to the tool where the project actually lives, and what that connection can and cannot do.

By · Updated · 6 min read

On this page
  1. The short answer
  2. Why a plan in a chat does not hold
  3. How MCP works, in plain words
  4. Connect Claude
  5. Connect ChatGPT
  6. Other apps: use a personal access token
  7. What the AI can do inside OIO
  8. The limits, and why they are there
  9. A first session to try

The short answer

Claude and ChatGPT can both connect to outside tools through MCP, the Model Context Protocol. Once connected, the assistant can read your projects and write to them: create tickets with due dates, add to-dos, add notes, change statuses and post a status line the client sees. You add the connection once in the assistant's settings, sign in to the project tool, and choose whether it may only read or also change things.

In OIO this is on the Pro plan ($29 a month) and up. The setup takes about two minutes for Claude.

Why a plan in a chat does not hold

Ask an assistant to plan a job and you get a decent list: phases, deliverables, a rough schedule. Then the chat scrolls away. The list is not on anyone else's screen, and when the client moves the deadline the plan does not move with it.

So people copy the plan into their project tool by hand, which is the slow part. MCP removes the copying: the assistant writes the tickets itself, in the right project, with the dates, where the team and the client already look.

How MCP works, in plain words

MCP is an open standard for letting an AI assistant call tools in other software. The project tool runs an MCP server, which is a web address that answers a short list of named actions, such as "list projects" or "create a ticket". Each action has a description and a fixed set of inputs. The assistant reads that list, decides which action fits what you asked, and calls it.

Three things matter for you as a user:

  • The assistant only gets the actions the server offers. If there is no "delete" action, it cannot delete.
  • The connection signs in as you. It sees what you see in the app and nothing more.
  • What the assistant reads is sent to the assistant's provider (Anthropic for Claude, OpenAI for ChatGPT), the same as anything you paste into a chat.

Connect Claude

These steps match Claude's settings as of October 2026. Anthropic has moved the menu once already this year, so if a label differs, look for Connectors.

  1. In OIO, open Settings, then Connections. Copy the server address shown there.
  2. In Claude (web, desktop or the phone app), open Settings, then Connectors, and choose Add custom connector.
  3. Paste the server address, give it a name such as OIO, and press Add or Connect.
  4. Claude opens an OIO sign-in page. Sign in and press Allow. The screen shows which app is asking and where it will send you back.
  5. The connection now shows in OIO under Settings, Connections, marked as connected via OAuth. You can disconnect it there at any time.

No token is needed for Claude; the sign-in replaces it. Anthropic lists custom connectors on its Free, Pro and Max plans, with Free limited to one. On Team and Enterprise, an owner adds the connector for the organisation first, then each member signs in.

For Claude Code, the one-line command is on the same OIO settings page; run /mcp afterwards to sign in.

Connect ChatGPT

ChatGPT accepts custom MCP connectors through developer mode. As of October 2026 the path is Settings, then Apps and Connectors, then Advanced settings, where you switch on developer mode. After that you create a connector, paste the OIO server address, and choose OAuth as the sign-in method. ChatGPT then sends you to the same OIO sign-in and Allow screen as Claude.

Two caveats. OpenAI has changed which plans get developer mode, and whether write actions are allowed on each plan, more than once this year, and some workspaces need an admin to allow connectors first. Check OpenAI's current help page before you count on it. And our settings page documents Claude; ChatGPT uses the same open sign-in standard, so the flow is the same in shape, but its screens are OpenAI's to change.

Other apps: use a personal access token

Cursor and most other MCP clients take a token instead of a sign-in. In OIO:

  1. Open Settings, then Connections, and choose New connection.
  2. Name it after the app that will use it, for example "Cursor on my laptop".
  3. Leave "Can make changes" on if the AI should mark tickets done and add notes, to-dos and tickets. Turn it off for read only.
  4. Pick when it expires: 30 days, 90 days, 6 months or a year.
  5. Copy the token. OIO shows it once and only keeps a scrambled copy, so treat it like a password.

The settings page has ready-made config snippets for Claude Code, Cursor and Claude Desktop. You can keep up to ten live connections and revoke any of them with one click.

What the AI can do inside OIO

The OIO server offers ten actions. That is the whole list:

ActionWhat it doesChanges data
List projectsYour projects, ones shared with you and ones you follow as a clientNo
List ticketsThe tickets on one project, with status, dates and to-do countsNo
Get a ticketOne ticket's note, to-dos and datesNo
Create a ticketA new ticket with a name and a due date (tomorrow if none is given)Yes
Add a to-doOne checklist item at the end of a ticket's listYes
Tick off a to-doMarks a to-do done, or open againYes
Add to a ticket's noteAppends a dated paragraph; never edits what is thereYes
Set ticket statusNew, in progress, problem, on hold, done or infoYes
Post a status updateA short line on the project, visible to the client on their status linkYes
Get the client status linkThe existing password-protected link for the project's clientNo

A ticket the AI creates lands on the project's kanban and timeline like any other, with today as the start and the due date as the end. The client sees ticket names and statuses on their status link, so ask the AI to name tickets the way you would want the client to read them. Notes stay internal.

The limits, and why they are there

We kept the server small on purpose. An assistant that misreads one instruction should not be able to do damage you cannot undo.

  • It cannot create projects or clients. You create the project in OIO, then point the AI at it.
  • It cannot delete anything, share anything, touch quotes, invoices or money, or change settings.
  • It can only change projects you own. Projects shared with you, and projects you follow as a client, are read only through the connection.
  • Every change it makes appears in your Logs with the connection's name, and every change can be reversed in the app.
  • Each connection is limited to 60 calls a minute and 2,000 a day.

Ticket notes are sometimes written by clients or freelancers, and a note could contain text that looks like an instruction to an AI. OIO marks that content as untrusted for the assistant, but no marking is perfect, so read what the AI says it did.

A first session to try

Create a project in OIO for a real job, then paste the client's brief into the chat and ask: "Read the OIO project called Riverside Tower and make a ticket for each deliverable in this brief, due on the dates it gives. Add the steps for each as to-dos. Check the existing tickets first so you do not duplicate any." Then open the project in OIO, on desktop or on your phone, and look at the timeline.

If you price the job before you plan it, the quote template and the day rate calculator cover that part. For studio-specific uses, see AI workflows for 3D and archviz studios, and for what a chat on its own leaves out, can AI replace project management software. How studios run jobs in OIO is on OIO for studios, and plans are on pricing.

Questions

Can ChatGPT or Claude manage my projects for me?

They can do the typing: turn a brief into tickets, add to-dos, update statuses and write a status line. They need a project tool to write into, and you still decide what goes in. In OIO you create the project; the AI fills and updates it.

Is it safe to give an AI access to my project data?

In OIO the connection signs in as you and the database applies the same rules as in the app, so it cannot see other people's data. It cannot delete, share or touch money. What it reads goes to the AI provider, the same as pasting it into a chat.

Which OIO plan includes the AI connection?

Pro and up. On lower plans a connection works only for the free Blender render link. Claude's custom connectors are available on Anthropic's own Free plan too, limited to one connector.

OIO is project management for small studios: one page per client job, tickets on a kanban and a timeline, freelancers on a ticket link and a status link for the client. On Pro and up, Claude and other MCP apps can read and update it for you. Create a free account.

More guides