Connect your AI to Nudje in two steps.
Nudje has its own MCP. Paste one address into your AI, sign in to Nudje, and your AI can tell you the state of your integration, then propose the events to send. Your key never goes through it.
- get_setup_status
- get_tracking_coverage
- get_tracking_playbook
- plan_tracking
- get_integration_brief
- prepare_env
- send_test_event
- check_events
Two steps, then ask it.
Add Nudje to your AI
In Claude, open Settings, then Connectors, then Add custom connector. Name it Nudje, paste this address, then press Connect.
A drawing of the step, not a screenshot.
Sign in to Nudje and allow it
Your AI opens a page of Nudje. You choose the workspace, you read exactly what it can and cannot do, and you press Allow.
It can
- See whether your setup is finished and which sources send events
- Read the catalogue of events Nudje understands
- Read the events Nudje received, to check that a test event arrived
It cannot
- Send, change or delete anything in your workspace
- See any API key, secret or password
Example. A drawing, not a screenshot: the AI app and the workspace are invented, and yours show your own.
Use an AI that can open your code, so it can read it and write the events with your yes. A chat with no access to your code can still tell you the state of your workspace.
Use Nudje and tell me the state of my integration.
Then ask it that. It starts from what I already hold about your workspace.From the state of your setup to a checked event.
This is what the eight tools hand your AI, in order. I have only run this with Claude on our own workspace, so read it as what the connection makes possible, not as a result.
It tells you what is set up
Your AI asks me what your workspace holds: what is connected, which events arrived, whether they attach to your billing customers, which moments are fed and which are empty.
It reads your code, and asks you
I hand your AI a protocol. Your AI reads your code, I never do. It asks what matters in your product, then shows a table: each event, with the file and the line.
Event, and where it would go
You say yes
Nothing is edited before your yes to that table.
Your answer
Yes, go ahead.It writes the code and sends a test
Your AI writes the integration after your yes. The key stays between you and your own .env: it writes an empty line and gives you the link to copy it yourself.
.env.example
+ NUDJE_API_KEY=You paste the key in your own .env.
It checks that it arrived
Your AI reads on my side what came in, accepted or refused, with the reason in words.
With the reason in words when one is refused.
Example. The events, files and lines are invented: yours come from your own code.
What it can and cannot do.
It can
- See what is connected and what is missing
- Read event names, counts, dates and cleaned page paths
- Read which moments and signals are fed or empty
- For the events it checks, read the account each one was sent for
- Read the brief I wrote for a source
It cannot
- Write, change or delete anything in your Nudje workspace
- See your API key, or any secret
- See the values inside an event
- See your code: your AI reads it, I do not
- Edit your code: your AI does, and only after your yes
Security
- Read only
- Revocable in Settings
- Your key stays yours
- You choose the workspace
Every service that touches your data is named on the security page.
What you may still wonder.
A standard way for an AI to use a tool. The Nudje MCP is one address you give your AI. Through it, the AI can ask me about your workspace, with eight tools that only read.
It works with Claude today. Any client that supports a remote MCP server with sign in can use the same address. I have not tested Cursor or ChatGPT yet. To set up tracking, use an AI that can open your code, so it can read it and write the events with your yes.
Mostly names, counts and dates. One tool, the one that checks what arrived, can show the account each recent event was sent for, which is a billing customer id or an email. It cannot see the values inside an event, and it never sees your key.
Yes. In Nudje, open Settings, then Security, then AI connections, and revoke it. The AI stops on its next call. You can connect it again whenever you want.