MCP documentation
Once connected, your assistant gets something like a universal remote for Nirvai — a small, consistent set of buttons that reach every part of your account. This page explains what those buttons are, and the rules that keep them safe.
What your assistant can reach
The connection speaks MCP (Model Context Protocol) — the shared standard modern AI assistants use to work with an outside system. Through it, your assistant can act across your whole Nirvai account:
| Area | What your assistant can do |
|---|---|
| Memory | Recall and save long-term facts, so it remembers people and decisions across every chat. |
| Databases | Read and update your records; create new tables and spaces. |
| Tools & connections | Run your saved tools and hundreds of pre-built connections — with credentials resolved safely on Nirvai's side. |
| Agents | Build, edit and run your Nirvai agents, and read back what they replied. |
| Live apps | Build interactive dashboards on your real data, test them, and publish. |
| Automations | Draft and trigger workflows. |
| Web & docs | Search the web, and answer from Nirvai's official documentation. |
Your assistant works with references to your things (this agent, that table), never by holding your data. When it needs the detail, it asks Nirvai for it just-in-time. So even a huge database never has to be pasted into a chat.
How it acts safely
Three rules make the connection safe to hand to any assistant you trust:
- It proposes, you approve. Anything that creates or changes something opens a small review page with a single Create/Connect button. Nothing happens until you click.
- Secrets stay yours. Your assistant can propose a login but can never complete one — you type the secret on Nirvai's own secure page. It never passes through the assistant.
- Scopes gate everything. The token you minted decides what's allowed:
readandrunare confirmation-free;writeandcreateare what let it change and build.
Claude CodeReady to connect- TypeAPI key
- You enterthe secret key
Work is grouped into sessions
Your assistant does related work under one session — a single named thread. Everything in a session accumulates in one place: you see it as one run in the Activity Feed, and your assistant can pick up exactly where it left off. Files it produces are synced to that session, so a later step (or a later day) continues cleanly.
Built-in commands
Type /nirvai: in your assistant to see the shortcuts — connect, context, remember, recall, new_agent, new_database, run_agent, activity, help and more. Each one kicks off the right sequence for you, so you don't have to know the individual actions.