Connect ChatGPT to your Koltrix inbox with a custom connector
Create a custom connector, choose OAuth and approve Koltrix in ChatGPT, learn why it gets search and fetch tools, and see what to ask first once connected.

On this page(10 sections)
ChatGPT can work with your Koltrix inbox through a custom connector, and the setup is shorter than most people expect: a URL, an authentication choice, and a sign-in.
This post walks through it, explains why ChatGPT sees two extra tools that Claude does not need, and lists what to try first. If you are comparing assistants, connecting Claude is nearly the same process.
Before you start
You need a Koltrix workspace with a mailbox you can open, and a ChatGPT plan that allows custom connectors. Custom connectors are a feature of ChatGPT's own apps and connectors settings, and which plans get them, and whether you must switch on a developer mode first, is decided by OpenAI and has changed over time. If you cannot find the option, that is the first thing to check.
The Koltrix server URL, the same one every assistant uses:
https://mcp.koltrix.com/mcp
Step 1: Create the connector
- In ChatGPT, open Settings and find Apps & Connectors. Under Advanced settings, turn on Developer mode if your plan requires it for custom connectors.
- Choose Create (or Add custom connector) and name it
Koltrix. - Paste
https://mcp.koltrix.com/mcpas the URL. - Choose OAuth for authentication.
Menu names differ between ChatGPT plans and change from time to time. Look for the words "connectors" or "apps", and expect the labels above to drift. The Koltrix docs page for AI assistants is kept current when the names change.
Step 2: Sign in and approve
ChatGPT sends you to a Koltrix approval page. It shows the app's name, the workspace to connect, and the permissions requested, in plain words:
- Read: search and read mail, list mailboxes and labels.
- Organise: add and remove labels, archive, mark read or unread, star.
- Draft: save drafts to your Drafts folder.
- Send: only appears if an admin has allowed it for the workspace, as an unticked box.
Click Approve. The connection belongs to you, in that one workspace, and uses standard OAuth sign-in, so you never give ChatGPT your Koltrix password. If you later want it gone, you can revoke it from your Koltrix settings, and it stops working immediately. Reviewing and revoking AI app connections covers the habit of checking that list now and then.
Why ChatGPT gets search and fetch
Koltrix's server offers a full set of tools. On top of those, it also offers two tools named search and fetch, because ChatGPT's connectors and deep research expect exactly that pair.
| Tool | Input | Returns |
|---|---|---|
search |
A query | Matching conversations as {id, title, url} |
fetch |
The id from search |
One conversation as {id, title, text, url, metadata} |
The search tool understands Koltrix's operators, so a query can include things like from:, to:, subject:, label:, has:attachment, is:unread, in:sent, before: and after:. You rarely type these yourself. You describe what you want and the assistant writes the query, but it helps to know that "unread invoices from last month with an attachment" has a precise form underneath.
The richer tools are there as well: search_mail with filters for folder, label, sender and unread, list_threads with a category filter, get_thread for a whole conversation, and the organizing and drafting tools. ChatGPT can use whichever fits the question.
Step 3: Use it in a chat
Start a conversation, enable the Koltrix connector for it, and ask a read-only question first:
Which conversations in my inbox are waiting on a reply from me?
If you are using deep research, you can ask a larger question that spans months of mail:
Look through our threads with Acme this year and list every commitment we made, with dates.
Deep research tends to run many searches and reads. That is a good use of read access, and it is also a reason to be thoughtful about scope: the assistant can only see what you can open, but you may not want a long research session pulling in a mailbox that is full of unrelated customer conversations. Ask about a sender, a label or a date range rather than "everything".
First things to try
| Goal | Ask |
|---|---|
| Orientation | "How many unread conversations do I have, by category?" |
| Retrieval | "Find the last thread where Priya mentioned the renewal date." |
| Cleanup | "Label receipts from last week 'Receipts' and archive them." |
| Drafting | "Draft a reply to the newest message from Dana saying we ship Friday." |
For the cleanup example, a label name that does not exist yet is created, so double-check the spelling. Everything it organizes can be undone from Koltrix.
For the draft example, the result is a draft in your Drafts folder plus a link that opens it in Koltrix. With sending off, which is the default, you review and send from Koltrix. Drafting replies with Claude and Koltrix has patterns for getting drafts that sound like you.
Sending from ChatGPT
By default, ChatGPT can read, search, organize and draft. Sending is off. It becomes possible only when three things are true: a workspace owner or admin turns it on, you tick the send box when you connect, and for each message ChatGPT shows you the recipients, subject and full text and you say yes in the conversation. Koltrix also checks that the recipients and subject the assistant names match the actual draft, and it applies hard limits: at most 10 recipients, 20 sends a day per connection and 50 a day per workspace through assistants, with no Bcc and no attachments.
The send call is also marked as irreversible, so ChatGPT is expected to ask for its own approval before running it, on top of Koltrix's checks. The details are in safe sending from Claude: the confirmation flow, and the same logic applies in ChatGPT.
What ChatGPT does with your mail
When you ask about your mail, the parts the assistant reads are sent to OpenAI and handled under OpenAI's terms. Koltrix sends only what a tool call asks for, and it does not receive your conversations with ChatGPT, only the tool calls it makes. Read what your AI provider sees when you query your inbox before connecting a mailbox with sensitive threads.
If it does not connect
- The connector cannot be created. Check that your plan supports custom connectors and that developer mode is on if required.
- The approval page says a redirect address is not allowed. The app tried to send you back to an address that is not
https(or a local address). Update the app or report it to its maker. - ChatGPT cannot find a message you can see in Koltrix. You may have approved a different workspace. Connect again and pick the right one.
- Connecting from a work network fails. The network has to let your app reach
mcp.koltrix.comand your browser reachapp.koltrix.com.
More fixes are in troubleshooting a Koltrix MCP connection.
Key takeaways
- Create a custom connector named Koltrix with the URL
https://mcp.koltrix.com/mcpand OAuth authentication, then approve it on Koltrix. - ChatGPT uses
searchandfetch, which Koltrix provides alongside its full tool set. - Out of the box, ChatGPT reads, searches, organizes and drafts. Sending needs an admin switch, your opt-in, and your yes on each message.
- Scope big research questions by sender, label or date, and remember the assistant's provider receives what it reads.
- The permanent reference is the AI assistants page in the docs, and there is an overview at koltrix.com/mcp.
Start with Koltrix
Your domain, one inbox, and an API that sends.
A team inbox where AI sorts and drafts (nothing is sent without your click), plus the transactional API and SMTP relay your product sends with. 7 days free, no card.


