How to connect Claude or ChatGPT to your own data
A connector lets an assistant read one app on your behalf. This page explains what you are actually adding, how sign-in works, and how to keep the access as small as you can.
At a glance
| Question | Answer |
|---|---|
| What you add | A server address, or a ready-made connector in the app |
| How you sign in | Approve in your browser, or paste a key you created |
| Who sees the data | The assistant you approved, for the scopes you picked |
| Can it change things | Only if you allow writes |
| How to stop it | Remove the connector, or delete the key |
MCP or API: what is the difference
An API is a way for software to talk to software. MCP is a shared shape on top, so any assistant can use any tool without custom code for each pair.
So MCP does not replace an API. It wraps one, and it means the same connection works in Claude, ChatGPT, Cursor and others.
How sign-in works
Claude and ChatGPT both let you add a connector and then sign in through your browser. You see a screen listing what is being asked for, and you press Allow.
Cursor and code tools usually want a key instead. Make one key per assistant, and give it the smallest scopes that answer your question. Reading costs and reading renewals do not need write access.
- Start with read-only, and add writes later if you miss them.
- Name each key after the tool using it.
- Delete a key the moment you stop using that tool.
What a connector should never ask for
No connector needs your bank login, your card number or your account password. If one asks, stop.
In LemSubs, an assistant sees the subscription list you built. Card details and passwords are never sent, and your audit log shows every call, the tool used and the time.
Questions people ask
What is the difference between MCP and an API?
An API is how two pieces of software talk. MCP is a shared shape on top of that, so any assistant can use any tool without custom code for each pairing.
How do I add an MCP server to ChatGPT or Claude?
Open the connectors section, add the server address, then sign in through your browser and approve the scopes. Code tools such as Cursor use a key you create instead.
Can I limit what an assistant is allowed to do?
Yes. Reads and writes are separate scopes, writes start switched off, and you can make one key per assistant so removing one does not affect the others.