Connect Claude to Koltrix, step by step
Add Koltrix as a custom connector in Claude, approve the permissions in plain words, and run your first read, organize and draft requests. One URL, no API key.

On this page(9 sections)
The whole setup is one URL and one approval screen. If you have used "Sign in with Google" to connect an app, you already know the shape of it.
This walkthrough connects Claude (on the web or in the desktop app) to a Koltrix workspace, tells you what you should see at each step, and covers the first things to ask once it works. If you use Claude Code instead, there is a command-line version in Connecting Cursor and Claude Code to Koltrix.
What you need first
- A Koltrix workspace with at least one mailbox you can open. The assistant gets exactly your access and no more.
- A Claude account that can add custom connectors. On Team and Enterprise plans, an owner may need to add the connector for the organization first. After that, each person connects their own Koltrix account.
- The server URL, which is the same for everyone:
https://mcp.koltrix.com/mcp
There is no API key to create. The first time Claude connects, it sends you to Koltrix to sign in and approve.
Step 1: Add the connector
In Claude, open Settings, then Connectors, and choose Add custom connector. Give it a name you will recognize, such as Koltrix, and paste the URL above.
Claude Desktop uses the same connectors as claude.ai. Add Koltrix in either app and it is available in both. You do not edit a configuration file.
Menu labels change between plans and releases, so if yours differs slightly, look for "connectors" or "integrations". The setup guide in the docs carries the current wording.
Step 2: Sign in and approve
Click Connect. Claude opens a Koltrix page in your browser. This page is the important one, because it is where you decide what Claude may do. It shows:
- the name of the app asking, and the website it will send you back to,
- the workspace to connect, if you belong to more than one,
- the permissions it asks for, in plain words,
- if an admin has allowed sending in that workspace, an unticked box labeled Also allow sending email.
Leave the sending box unticked for now unless you have a specific reason to tick it. We will come back to it. Click Approve, and Claude takes you back to the conversation.
The connection belongs to you, in that one workspace. To use a second workspace with Claude, connect again and choose the other workspace on the approval page.
The permissions, in plain words
| Permission | What it lets Claude do |
|---|---|
Read (mail.read) |
Search and read your mail, and list your mailboxes and labels |
Organise (mail.organize) |
Add and remove labels, archive, mark read or unread, and star. All of it can be undone in Koltrix |
Draft (mail.draft) |
Save drafts to your Drafts folder. On its own this cannot send them |
Send (mail.send) |
Send a draft that already exists, after showing you the message and getting your yes. Off unless an admin allows it and you opt in |
Apps ask for the first three by default. The fourth is never part of that default. If you want the long version of how these fit together, how OAuth and permissions work for email assistants goes through it.
Step 3: Turn Koltrix on in a chat
Connecting makes Koltrix available; it does not mean every chat uses it. Open a conversation and turn Koltrix on from the tools menu. From here, Claude can call Koltrix tools when your question needs them, and Claude usually shows you when it does.
Step 4: Ask something small first
Start with a read-only question so you can check that it sees what you expect:
What came in overnight that needs a reply from me?
Behind the scenes, Claude is likely to call get_inbox_summary, which returns counts by category, unread counts, and the conversations waiting for your reply. Then it may open a thread or two with get_thread to describe them. If the answer matches what you see in Koltrix, the connection is working.
Next, try something that searches:
Find the thread with Dana about the contract renewal and summarize it.
Then something that organizes, because it is the quickest way to feel what the assistant can do:
Label the receipts from the last week "Receipts" and archive them.
Two details matter here. If the label you name does not exist yet, Koltrix creates it, so a typo makes a new label rather than an error. Check the spelling, or look at your labels first with a quick "list my labels". And everything it just did can be undone from Koltrix, which is why organizing is a comfortable first write action.
Finally, ask for a draft:
Draft a reply to the latest message from Acme saying we'll ship on Friday.
The draft is saved in your Drafts folder, and Claude gives you a link that opens it in Koltrix, where you can edit it. With sending off, which is the default, nothing leaves until you open that draft and click Send yourself.
What a good first session looks like
| Minute | You ask | What you check |
|---|---|---|
| 1 | "What needs a reply?" | The list matches Koltrix |
| 3 | "Summarize the thread with Dana" | The summary matches the thread |
| 5 | "Label and archive last week's receipts" | Labels applied, Inbox shorter |
| 8 | "Draft a reply to Acme" | The draft is in Drafts and reads right |
If all four feel right, you have a working setup and a sense of where the assistant is reliable.
What Claude cannot see or do
Knowing the edges makes the tool easier to trust.
- It sees only the mailboxes you can open in Koltrix. A teammate's mailbox you have no access to stays invisible to Claude too.
- It cannot delete mail or forward it. There is no tool for either.
- It does not download attachments. Attachments are listed by name.
- It cannot send anything unless sending was allowed by an admin and you opted in. Even then it must show you the recipients, subject and full message and get your yes.
Koltrix's own AI, the agents that sort and summarize inside the app, is separate. It never sends without a person clicking Send.
If something does not work
The most common problems are boring and quick to fix:
- Claude says it is not authorized or asks you to sign in again. The connection was revoked, expired after 30 days of not being used, or the workspace turned assistants off. Reconnect from Claude.
- It cannot find a message you can see. Check that you approved the right workspace. Koltrix only searches the workspace the connection belongs to.
- Reading works but labeling or drafting fails. You did not approve that permission. Revoke the connection and connect again, approving it.
The longer list, with fixes, is in troubleshooting a Koltrix MCP connection.
Key takeaways
- Setup is one URL,
https://mcp.koltrix.com/mcp, plus a sign-in and approval on Koltrix. No API key. - Read the permissions on the approval page; leave the sending box unticked until you have a reason to use it.
- Start with a read-only question, then organize, then draft. Each step shows you something about how the assistant behaves.
- The assistant has your access and no more, and everything it organizes can be undone.
- For the permanent reference, use the AI assistants page in the docs, and see what the integration offers on koltrix.com.
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.


