In short
- Dagomi is a tool inside ChatGPT and Claude: check a company, search companies, add new companies, fetch your own knowledge.
- A key belongs to one person. Scopes are enabled individually — the default is the smallest scope, revenue and writing are off.
- Every call is logged. The key appears exactly once in plain text; after that only a check value is stored on our side.
What the connection can do
Four tools are available to the assistant. Check company: do we know this business already, and what do we know? Search companies: who in our database matches a description? Add company: record a new company with a note. Fetch own knowledge: your dictated thoughts, shared summaries from the team and your company’s FAQ.
The example it was built for: “Find me ten machine builders in East Westphalia with over 100 employees. Check each one against our records — and add the new ones.” Without the connection you would get a list half full of companies you have served for years.
An open standard, not a plugin
Dagomi speaks the Model Context Protocol (MCP) — the open standard through which assistants like Claude and ChatGPT use tools. It is not a plugin of a single vendor: what works in Claude today works in the next assistant that speaks the standard, without changing anything in Dagomi.
One key per person, scopes set individually
An access key belongs to a person, not to the company. It carries a name (“Claude laptop”, “ChatGPT sales”) and a list of scopes: company basics (default), notes and history, revenue data, own knowledge and company knowledge, create and append. Whatever is not enabled, the server does not hand out — even if the assistant asks for it.
Two scopes are deliberately off until you switch them on: revenue data, because those are the most confidential figures, and writing, because an assistant should not create anything unless you have decided so. Writing means creating and appending — nothing changed, nothing deleted.
Own knowledge — without customer notes
The “Own knowledge” scope delivers your thoughts on stance, experience and expertise, shared summaries from the team and the FAQ. Whatever belongs to a company or a person never leaves through here — neither customer notes nor conversation histories. That lets a weekly task in Claude write posts that sound like you, without a single customer name finding its way outside.
Where the data goes — and why that is a conscious decision
Whoever connects Dagomi to ChatGPT sends customer data to OpenAI; with Claude, to Anthropic. That is the point of the feature, but it has to be a decision, not a side effect. So before a key is created, it says in plain language what goes out, and the default is the smallest scope. Every call is logged: tool, time, key. “You can set what goes out” would be a mere claim without that log.
The key itself appears exactly once in plain text. After that, only a check value is stored on our side, from which the key cannot be recovered. Lost means: create a new one. That is inconvenient and right.
How to set it up
-
Set up access
In Setup → “Connect AI assistants”, create an access key: give it a name, choose scopes. Copy the key immediately — it is shown only once.
-
Add a connector in Claude
Claude: Settings → Connectors → Add → Add custom connector. Choose any name, enter the server address from Dagomi as the URL.
-
Authentication: “None”
That sounds wrong but is right — this option covers servers that use a key instead of a sign-in. With “Always required”, Claude looks for an authorisation server that does not exist here.
-
Enter the key as a header
Under “Additional request headers” add one row: Authorization → Bearer plus your key. If the assistant does not allow that name, use X-Dagomi-Key and enter only the key.
-
Ask
Save, then ask in chat: “Do we already have company XY in our records?” In ChatGPT it works the same way via the connector settings.
-
Check the log
Back in Dagomi you see on the key when it was last used. If something is off, delete it — the assistant gets nothing from that moment on.
Good to know
- Access applies to the person who owns the key. The log shows their name — not “the company”.
- An assistant cannot change or delete anything through Dagomi. Even with the “Create & append” scope, only new companies and appended notes come into being.
- A deleted key is invalid immediately; a new one is created in a minute.
Frequently asked questions
Do I need to be able to code?
No. You copy an address and a key into your assistant’s settings — described step by step above. No code, no server of your own.
What does OpenAI or Anthropic see of my data?
Exactly what the assistant requests and what you have enabled in the key — nothing more. Without the “Revenue data” scope no revenue figures leave Dagomi, without “Notes & history” no conversation notes. And every call is in the log.
Can the assistant delete customers or change notes?
No. There is no tool for that. Creating and appending is the maximum — and even that only if you have switched the scope on.
Does it work with assistants other than Claude and ChatGPT?
With any assistant that speaks the MCP standard and can set a custom request header. The setup is tested and described for Claude; in ChatGPT it runs through the connector settings.
How do I revoke access?
Delete the key from the list — effective immediately. An assistant still using it gets a rejection from then on.