Most people have tried the same thing at least once: paste the day's tasks into ChatGPT, ask it to help prioritise, get back something that sounds sensible, and then type the result back into whatever app the tasks came from.
It works, in the sense that you get an answer. It just does not survive contact with a second day.
Why copy-and-paste planning falls apart
Three problems, and they compound.
The snapshot is stale before you finish pasting. You copy your list at 9am. By 11am two things are done, one has moved, and a new one has landed. The assistant is now reasoning about a version of your day that no longer exists, confidently.
Nothing it decides actually happens. The plan lives in the chat. Moving two tasks to tomorrow and adding a follow-up means going back to the app and doing it by hand, which is most of the work you were trying to avoid. In practice the plan gets skimmed and abandoned.
It has no context beyond what you pasted. It does not know which project the task belongs to, what the goal above it is, that you have missed the habit four days running, or that the deadline it is cheerfully ignoring is on a milestone. You can paste all of that too, but at some point you are reimplementing a database in a chat window.
The result is that the assistant is good at generic advice and bad at your actual week — not because the model is weak, but because it is guessing at a system it cannot see.
The other arrangement
The alternative is not "a chatbot inside your task manager". Almost every app has bolted one of those on, and it is the wrong way round: it asks you to leave the conversation you were already having and go and talk to a different, worse assistant that only knows about tasks.
The useful arrangement is the reverse. Your task manager exposes itself to the assistant you already use, so the assistant can read and change the real thing.
That is what a connector does. In EvyOS it is one URL — https://api.evyos.com/mcp — pasted into ChatGPT or Claude as a connector. You sign in, approve a consent screen, and from then on the assistant is looking at your account rather than a paragraph you pasted.
What changes when it can see the real data
The questions stop being hypothetical.
What have I got today, and what am I going to miss? It reads the overdue items, what is due, the habits still open and the calendar in a single call, and can tell you the honest answer — that the six things due today do not fit, and which two are load-bearing.
Move the two smallest to tomorrow and add a task to chase the invoice. It does it. The tasks move, the new one is created with the right project, priority and date, and when you open the app it is already true. No re-typing.
How is the rebrand actually going? It can walk the goal down to its projects and their milestones, rather than guessing from the four task titles you happened to paste.
I did an hour of Spanish. Logged against the skill, with the date and the minutes, from the conversation you were already in.
None of this is clever prompting. It is the difference between an assistant reasoning about a description of your system and one reasoning about your system.
Setting it up
It takes about two minutes and there is nothing to install.
- Add the connector. In ChatGPT or Claude, add a connector and give it
https://api.evyos.com/mcp. Both products discover everything else themselves — there is no API key to generate, store or paste. - Sign in and approve. You are sent to EvyOS to sign in as normal, then shown a consent screen naming the app that asked and what it wants. Approve it and you are returned to the assistant, connected.
- Ask it something. "What is overdue?" is a good first question, because the answer is checkable.
Connecting an assistant is a Pro feature, and Pro has a 14-day trial that does not ask for a card.
The things worth checking before you connect anything
Handing an AI client access to your real data is a genuine decision, and it is worth knowing what you are agreeing to — for any product, not just this one.
Where the credentials go. The connection uses OAuth 2.1 with PKCE, which means the client never sees your password or your session. What it holds is a token scoped to your account. An integration that asks you to paste a long-lived key into a chat interface is a worse deal than it looks.
Whether you can take it back. Connected apps are listed in Settings with when they were connected and when they were last used, and revoking one invalidates every token issued under it immediately. "Last used" is the useful column: it is how you notice the app you experimented with in March and forgot about.
What it can destroy. The connector tells the model explicitly that deleting is not the same as finishing, and to prefer completing a task over removing it. Deletions are also soft, so something removed by mistake is not gone from the database. Ask any tool you connect what happens on the bad day, not the good one.
What it does not fix
It will not decide what matters. An assistant with full access to your task list is very good at telling you that the day does not fit and quite bad at telling you which half to drop, because that is a judgement about your life rather than your calendar.
It also does not replace writing things down. The assistant is only as useful as the system underneath it: if your tasks have no projects, no dates and no goals above them, it can read all of that emptiness perfectly and still have nothing to say.
The connector removes the transcription. The thinking is still yours.
Going further
Run EvyOS from ChatGPT and Claude covers what the assistant can reach and what it looks like in use. If you want the technical side — transport, scopes, the full tool catalogue, how the tokens are stored — that is on the MCP server page, and what MCP is and why your task manager should have one is the background reading.