Connect — API, Code, Cursor, Lovable

Connect non-OAuth AI clients with a role-scoped access token minted in console.kula.digital, which generates the ready-to-paste config for each platform.

There are two ways to connect, depending on what your client supports — and both credentials are role-scoped and created in console.kula.digital, so access stays under your control. There's no open, anonymous endpoint.

Your client	Method	See
Claude Desktop / claude.ai	Custom connector (OAuth) with a connect link — one click from an invite, or pasted by hand	Connect Claude
ChatGPT (Plugins / MCP Servers)	Custom MCP server (OAuth) with the same connect link	Connect Claude
Claude API, Claude Code, Cursor, Lovable, custom agents	Access token (bearer)	this page

This page covers the access-token method, for any client that connects with a bearer token rather than the OAuth custom-connector flow.

Get an access token

1. Sign in to console.kula.digital and open Tokens → Mint a token. (Connectors → New invite is the other path — that mints an OAuth connect link for the Claude/ChatGPT custom-connector flow, not a bearer token.)

2. Choose the permission level for the integration — see permission levels.

3. Pick your platform. The app generates the ready-to-paste connection config for that platform (and the token itself, shown once). Copy it.

The token is a role-scoped credential, exactly like a connect link — it just travels in an Authorization: Bearer header instead of in the URL. The endpoint it points at is:

https://mcp.kula.digital/mcp        (Authorization: Bearer <token>)

Use the per-platform config the app gives you — the snippets below show the shape so you know what to expect, but the app fills in your token and the exact format for your client.

Claude API (programmatic)

Pass the token as authorization_token on the MCP server entry.

from anthropic import Anthropic

client = Anthropic()

response = client.messages.create(
    model="claude-opus-4-8",  # or whichever Claude model is current when you read this
    max_tokens=4096,
    mcp_servers=[{
        "type": "url",
        "url": "https://mcp.kula.digital/mcp",
        "name": "kula-intelligence",
        "authorization_token": "YOUR_TOKEN_HERE",
    }],
    messages=[{
        "role": "user",
        "content": "Which instructor has the highest affinity with member M123?",
    }],
)

print(response.content)

Claude Code (CLI)

claude mcp add --transport http kula-intelligence https://mcp.kula.digital/mcp \
  --header "Authorization: Bearer YOUR_TOKEN_HERE"

Verify with claude mcp list, then use your studio data in any session.

Cursor

Add an MCP server in Cursor's settings with the URL above and an Authorization: Bearer YOUR_TOKEN_HERE header. Create the token at the operations or admin level for most work on member data — see permission levels.

Lovable

Add Kula as an MCP tool source with the URL and bearer header. For dashboards and embeds that show trends without naming individuals, create the token at the analytics level so no personal data is ever returned.

Any other MCP-compatible client

The server speaks standard MCP over streamable HTTP. Point the client at https://mcp.kula.digital/mcp with an Authorization: Bearer <token> header. The bare endpoint requires a valid role-scoped token — without one it returns 401, so there's no anonymous access.

Verify the connection

Ask the client:

"List the tools you can use from Kula Intelligence."

You should get the full tool list back — scoped to the token's level.

Good practice

• One token per integration, so usage and revocation are independent.

• Lowest workable level — see permission levels.

• Never commit a token. Store it in your platform's secret manager, and revoke it from console.kula.digital if it leaks.