Connect an AI agent

This router is agent-native: assistants can discover it, sign in as you, and call your models without anyone pasting docs into a prompt.

1. MCP server (recommended)

Works with Claude, Cursor, ChatGPT connectors and anything else speaking MCP over Streamable HTTP. You sign in with your account — no API key to paste.

Server URL

https://your-router.example/mcp

Config file snippet

{
  "mcpServers": {
    "llm-router": {
      "type": "http",
      "url": "https://your-router.example/mcp"
    }
  }
}

Tools exposed

list_models — every model name this router serves
chat — run a prompt through one of them
usage_summary — your own request counts, latency and tokens

2. OpenAI-compatible API

For agent frameworks that take a base URL and key — LangChain, LlamaIndex, the OpenAI SDKs, and anything else OpenAI-shaped.

Base URL

https://your-router.example/api/public/v1

from openai import OpenAI

client = OpenAI(
    base_url="https://your-router.example/api/public/v1",
    api_key="rk_live_...",  # minted on your dashboard
)

resp = client.chat.completions.create(
    model="qwen3-30b",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

3. Discovery files

Crawlers and agents read these automatically to learn what this service offers.

/llms.txtplain-language summary for AI clients /openapi.jsonOpenAPI 3.1 spec, models generated live /.well-known/ai-plugin.jsonplugin manifest

Access model: MCP callers sign in with their own account and approve the client; REST callers use their own router key. Either way, upstream provider credentials stay on the server and every call is logged against the account that made it.