Skip to main content

Connecting an AI assistant

Nydli can let your own AI assistant work in a workspace: Claude, ChatGPT, or anything else that speaks MCP. It's off by default, and nothing about it happens unless you switch it on.

Nydli has no AI of its own. We never send what you write anywhere to be analysed, and no assistant sees anything until you connect one yourself.

Turning it on​

Settings → Data → MCP server. Only the workspace Owner can, and switching it on asks for your account password, because this is the one switch that lets something outside Nydli reach in. Everyone in the workspace can see when it's on.

Two modes:

  • Read only. The assistant can look up your tasks, categories, labels and tracked time.
  • Read and write. It can also create and change tasks, categories and labels, and archive them.

It can never delete anything. We didn't build it a tool for deleting, or one for the bin, or one for writing time entries. Archiving is reversible with no time limit, which is exactly why we allow that one.

Changes an assistant makes look like they came from the person who connected it, and they show in the Activity log that way.

What the assistant can reach​

Whatever you can reach, and no more. A connection inherits your role and access level, so a Limited member's assistant sees a Limited member's tasks. A frozen or archived workspace stays read-only for an assistant too.

Connecting​

The same page gives you a server address. Paste it into your assistant's connector or MCP settings, and it sends you to Nydli to sign in and asks permission, naming the site it's connecting from. Approve it and it works.

Connected assistants on the same page lists what's connected and when you last used it. Revoke cuts one off straight away, everywhere it could reach. The Owner sees every connection in the workspace; everybody else sees their own.