MCP

What it can do

Call a model, make a key with a monthly limit, show this week's earnings. Every Management API operation is a tool, with the same permissions and the same activity log.

A browser sign-in can do what an agent key can do. It cannot add wallets or create more agent keys. Disconnect it on API keys and it stops working.

Cursor and Claude

Home has an Add to Claude button. It opens Claude with Klinn already filled in, and Claude asks you to allow it. Or run the command. Cursor takes the same URL with no header. There is no key to copy.

claude mcp add --transport http klinn https://app.klinn.dev/api/v1/mcp --scope user

{
  "mcpServers": {
    "klinn": { "url": "https://app.klinn.dev/api/v1/mcp" }
  }
}

Claude Desktop uses npx mcp-remote https://app.klinn.dev/api/v1/mcp with no Authorization header. It opens the browser the same way.

Codex and other agents

These still take an agent key from the console, on Home or API keys. Codex reads it from an environment variable.

[mcp_servers.klinn]
url = "https://app.klinn.dev/api/v1/mcp"
bearer_token_env_var = "KLINN_MCP_KEY"

Gateways

An agent can set up a gateway too: fallbacks, a cache, limits, or a decisions classifier with its questions, and attach it to a key. The tools are list_gateways, get_gateway, create_gateway, update_gateway, delete_gateway, gateway_stats, gateway_decisions (a decision gateway's answers per question) and update_key, which sets or clears a key's default gateway. They take the same config as the Management API.

What it does not do

The server is stateless HTTP. It does not sign in with a wallet or pay over x402 for you, so agents that hold a wallet use the API directly.