Answers
Can ChatGPT change my plan, or only read it?
Two things have to be true, and almost everything written about this mentions only the first. The software has to publish actions that change things, which Empiraa GPS does. And your ChatGPT plan has to be one that is allowed to use them, which not every plan is. OpenAI has write actions through MCP in beta on Business, Enterprise and Edu; on Pro you can connect a server with read permissions, which means reading only.
OpenAI behaviour checked against their own help centre on 15 October 2026.
- Set up with your plan
- Unlimited users
- 30-day pilot
Why the answer keeps coming out wrong
Everyone answers half the question.
The vendor tells you what their server can do. OpenAI tells you what ChatGPT can do. Nobody puts the two together, and the overlap is where the actual answer lives.
Where the confusion comes from
A vendor says "we have an MCP"
True and incomplete. It tells you actions were published. It does not tell you whether those actions change anything, and it certainly does not tell you whether your assistant is permitted to call the ones that do.
The plan tiers are not obvious
The ability to write through a connector is not a toggle you find by looking harder. It is attached to the subscription. The same software, connected to the same account, behaves differently on two ChatGPT plans.
It reads fine on a phone and then does nothing
Custom MCP connectors in ChatGPT are web only. People test on a laptop, switch to a phone the next day, and conclude the connection broke.
The two gates
Both have to be open before anything changes.
How it works
Gate one: did the software publish actions that write?
This is the vendor's half. An MCP server that only looks things up is a search box with extra steps, and plenty of them ship that way. Empiraa GPS published writing actions in 1.22, so an assistant can record progress against a measure, create an action, update an objective or add a note, not merely read them back to you.- Ask any vendor for an example of a change made through their server
- "We have an MCP" is not the same claim as "it can write"
- GPS writes: progress, actions, objectives and notes
Gate two: is your ChatGPT plan allowed to use them?
This is OpenAI's half, and it is the one that surprises people. Their help centre says full MCP support including modify and write actions is rolling out in beta to Business, Enterprise and Edu plans. Pro users can connect MCP servers with read and fetch permissions in developer mode, which is reading. OpenAI's own built-in apps are search-only and do not write at all.- Business, Enterprise and Edu: write actions, in beta
- Pro: connect in developer mode, read and fetch only
- Web only. Custom connectors do not run on mobile.
And then the confirmations sit on top
Even with both gates open, a write is not silent. OpenAI say ChatGPT may ask for confirmation depending on the app's permissions, and that especially risky actions can be blocked rather than offered for approval at all. Administrators are shown a risk warning when they enable an app that can modify data. That is the right behaviour, and it is worth knowing before you conclude something is broken.What GPS does on its own side of the line
Independent of which assistant is connected, the plan stays yours to govern. The connection acts as the person asking, so it reaches what they already reach and nothing more. Every change made this way is labelled as made through an assistant. Bulk changes are previewed before they save. Most changes can be undone for seven days. And an administrator sets assistants to Off, View only or Full for the whole organisation.What to check before you blame the connection
Five things that explain almost every "it will not change anything".
In order, cheapest first. Most reports of a broken connector are one of the first two.
The diagnostic
Which ChatGPT plan is this?
Pro connects and reads. Business, Enterprise and Edu are where write actions live, and they are in beta rather than generally available.
Are you on the web?
Custom MCP connectors in ChatGPT are web only. Nothing you do on a phone will change that today.
Is developer mode on?
It is what allows a custom MCP connector at all. On Business plans an administrator enables it, and publishing an app to the workspace is an admin or owner action.
What access did your admin set?
In GPS, View only is a complete and sufficient explanation for an assistant that reads your plan perfectly and refuses to touch it.
Did a confirmation go unanswered?
A write may be presented for approval rather than performed. An unanswered prompt looks identical to a failure.
Is the change actually permitted?
The published list is the boundary. If nothing in it changes the thing you asked about, no amount of rephrasing will reach it.
Questions
Questions people ask about ChatGPT and their plan
Can ChatGPT change my plan, or only read it?
It can change it when two conditions are met: the software has published actions that write, and your ChatGPT plan is one that may call them. Empiraa GPS publishes writing actions. On OpenAI’s side, full MCP support including modify and write actions is in beta on Business, Enterprise and Edu plans, while Pro users can connect a server with read and fetch permissions, which means reading only.
Why can ChatGPT read my plan but not update it?
Three likely reasons, in order of how often they turn out to be the cause. Your ChatGPT plan allows reading but not writing, which is the case on Pro. Or an administrator has set assistants to View only in GPS. Or a confirmation prompt is waiting and has not been answered, which looks the same as a refusal.
Does this work on the ChatGPT phone app?
Not for custom MCP connectors. OpenAI say the feature is available on web and not on mobile, so a connection that works on a laptop will not be there on a phone. This is a current limitation rather than a setting you can change.
Do I need developer mode switched on?
Yes, for a custom MCP connector. Developer mode is what lets an organisation connect, test and deploy MCP apps in ChatGPT. On Business plans an administrator enables it, and publishing an app for the whole workspace is an admin or owner action rather than something each person does.
Is a change made by ChatGPT marked as made by an assistant?
In Empiraa GPS, yes. Every change made through an assistant is labelled as such, so you can tell later whether a person or an assistant recorded it. Bulk changes are previewed before saving, and most changes can be undone for seven days.
Can I let ChatGPT read the plan but never change it?
Yes, and it is the setting most organisations start on. An administrator sets assistants to Off, View only or Full for the whole organisation. View only lets anyone ask what is off track or who owns a goal while leaving the plan itself untouched.
Does the same answer apply to Claude?
The GPS half is identical, because the server is the same for any assistant that speaks the standard. The assistant half differs, since each vendor decides which of its plans and modes may call actions that write, so check the one you use rather than assuming they match.
Working out whether to switch it on? Ask ANI and get a straight answer.
Set the access first, then connect it.
Start free, choose Off, View only or Full for the organisation, and connect Claude or ChatGPT in under a minute.