Connecting to mente

There are three ways into mente, and they all reach the same place: the same tools, the same data, the same permissions. Pick the one that matches who — or what — is doing the asking.

You want to…	Use	What you need
Work with mente from Claude, Cursor, or another AI assistant	MCP	Nothing — you sign in through your browser
Run mente from your own code, a cron job, or CI	REST	A token
Drive mente by hand from a terminal	The command line	A browser login, or a token

Most people only ever do the first one.

Connect an AI assistant

mente is a remote MCP server. Your assistant connects to it over the internet — there is nothing to install and nothing to run on your machine. One address serves everyone:

https://api.usenous.ai/mcp

You never paste a token. The first time your client connects, it opens your browser, you sign in the way you always do, and you approve the connection. The client stores the credential itself and refreshes it silently from then on. That's OAuth 2.1, and every step of it — registering the client, signing in, staying signed in — happens without you thinking about it.

The setup instructions in the web app under Access & Keys say the same thing with copy buttons, if you'd rather click than read.

Claude Desktop

1. Click the search and tools icon (the magnifying glass) below the message box.
2. Scroll to the bottom and choose Add integrations.
3. Choose Add custom connector and paste the server address above.
4. Claude opens a browser window — sign in and approve.
5. mente's tools appear in the tools menu.

Claude Code

One command:

claude mcp add --transport http mente https://api.usenous.ai/mcp

Or add it to your project's .mcp.json:

{
  "mcpServers": {
    "mente": {
      "url": "https://api.usenous.ai/mcp"
    }
  }
}

Restart Claude Code and it opens a browser for sign-in. If it doesn't, run /mcp and authenticate from there.

Cursor

Settings → MCP → Add new global MCP server opens ~/.cursor/mcp.json. Paste this and save:

{
  "mcpServers": {
    "mente": {
      "url": "https://api.usenous.ai/mcp"
    }
  }
}

Cursor prompts you to authorize in your browser on first connect.

VS Code

Add .vscode/mcp.json in your project:

{
  "servers": {
    "mente": {
      "type": "http",
      "url": "https://api.usenous.ai/mcp"
    }
  }
}

Any other MCP client

Point it at the server address as a streamable HTTP transport. Clients discover where to sign in on their own, from https://api.usenous.ai/.well-known/oauth-protected-resource. Nothing else needs configuring.

The name is yours

mente in these config files is just a label — it's what the connection is called inside your client. Name it whatever you like.

Which workspace you're connected to

If you belong to more than one workspace, the one you're currently in when you authorize is the one the connection acts in, from then on. So switch to the right workspace in the web app first, then connect.

To move an existing connection to a different workspace, disconnect it and connect again from the workspace you want.

What your assistant sees

Not a wall of hundreds of tools. mente gives your assistant a handful of meta-tools and lets it look things up as it goes: it searches for a tool, reads what that tool needs, and runs it. The catalog it searches is yours — your permissions, your connected services, your workspace — so two people on the same team can see different things.

That means you don't have to tell it which tool to use. Ask for what you want.

Managing connected clients

Access & Keys in the web app lists every AI client connected to your account and lets you revoke any of them. Revoking takes effect immediately; the client will ask you to sign in again the next time it tries.

Call the API from your own code

Everything an assistant can do is an ordinary HTTP call. Find a tool, then run it:

curl https://api.usenous.ai/api/capabilities?query=task \
  -H "Authorization: Bearer $MENTE_TOKEN"

curl -X POST https://api.usenous.ai/api/capabilities/create_task \
  -H "Authorization: Bearer $MENTE_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"title":"Renew the Acme policy","priority":"high"}'

The token is an agent token — you create an agent in the web app and mint one from its page. See Agents, then How the API works for the catalog, the response envelope, error codes, and picking a workspace with the X-Mente-Org-Id header.

From the terminal

The mente command line does the same things by hand or from a script: mente login signs you in through the browser, mente tools search finds something, mente run executes it. For a scheduled job with no browser, an agent token in MENTE_TOKEN replaces the login.

The command line isn't published for general install yet.

Credentials at a glance

Credential	Who it's for	Works over
OAuth sign-in	You, in an AI client	MCP
Agent token (nous_pat_…)	Automation you run	MCP, REST, command line
API key (nk_…)	Older programmatic access, from Access & Keys	REST

Prefer an agent token over an API key for anything new: an agent has a name, its own permissions, and its own activity trail, so what it does is attributable and what it can reach is something you decide.

Whatever the credential, it's a secret. Keep it out of source control, and revoke rather than reuse when in doubt.

When something isn't working

The client keeps asking me to sign in. The connection was revoked, or you signed out everywhere. Reconnect from the client.

A tool I expected isn't there. The catalog is filtered to what you can reach. Connect the service under Connectors, or ask whoever owns the data to share it with you — then have your assistant look again.

It's working in the wrong workspace. See which workspace you're connected to.

It says something needs approval. That's by design — some actions are gated on a person saying yes. Approve it in the web app and ask again.

Next

Running automation rather than chatting? Set up an agent — it gets its own identity, its own token, and its own scope of access.