TMCP Gateway API reference

Connect a Gemini API Rotate or OpenRouter API Rotate tool, add multiple provider API keys to its pool, then point any OpenAI-compatible client at TMCP using your single agent API key. TMCP rotates across the pool round-robin, marks keys that return 429/quota errors as cooling down, retries the next key automatically, and returns the provider's original error only once every key is exhausted.

Each tool has its own dedicated base URLso you can drop it straight into any app's "OpenAI-compatible base URL" field — no model-name routing required. The endpoints under each base are /chat/completions, /responses, /embeddings, /models.

curl -X POST "https://your-domain.com/api/gemini/v1/chat/completions" \
  -H "Authorization: Bearer mcp_live_xxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "gemini-2.5-flash",
  "messages": [
    {
      "role": "user",
      "content": "Hello"
    }
  ]
}'

curl -X POST "https://your-domain.com/api/openrouter/v1/chat/completions" \
  -H "Authorization: Bearer mcp_live_xxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "openai/gpt-4o-mini",
  "messages": [
    {
      "role": "user",
      "content": "Hello"
    }
  ]
}'

Streaming is supported — pass "stream": true"and TMCP relays the provider's Server-Sent Events stream unchanged. List acceptable model ids with GET {base}/models. Standard parameters (temperature, top_p, max_tokens, …) are forwarded to the provider as-is.

Prefer a single endpoint? https://your-domain.com/api/v1 also works and auto-selects the provider from the model name (gemini*→ Gemini, otherwise OpenRouter).