API and MCP

Three steps to a working call

1. Create a free account — we e-mail you a sign-in link, there is no password to choose.
2. Generate an API key in the dashboard. You see the secret once; we only store its hash.
3. Send it as a bearer token:

curl -H "Authorization: Bearer llmo_sk_..." \
  "https://app.aimodeloverview.com/v1/deprecations?before=2026-12-31"

The response:

{
  "data": [
    {
      "provider": "anthropic",
      "model_id": "claude-opus-4-1-20250805",
      "name": "Claude Opus 4.1",
      "status": "deprecated",
      "deprecation_date": "2026-06-05",
      "retirement_date": "2026-08-05",
      "effective_date": "2026-08-05",
      "days_remaining": 9,
      "pricing": { "input_per_mtok_usd": 15, "output_per_mtok_usd": 75 }
    }
  ],
  "pagination": { "limit": 100, "returned": 1 }
}

Endpoints

Endpoint	What it answers
GET /v1/deprecations?before=2026-12-31	Models with an end date, soonest first — including days_remaining.
GET /v1/models?provider=anthropic&status=deprecated	Filterable, cursor-paginated model list.
GET /v1/models/{provider}/{model_id}	One model with its price history and source links.
GET /v1/providers	Every provider with its model count.
GET /v1/changes?since=2026-07-01	New models, status transitions and price changes — the feed for your own alerting.
GET /v1/me	Your organization, key and remaining quota.

Same key, from an AI assistant

The MCP server exposes the same dataset as tools, so Claude, an IDE or any MCP client can answer lifecycle questions directly. Point it at https://mcp.aimodeloverview.com/mcp:

{
  "mcpServers": {
    "ai-model-overview": {
      "url": "https://mcp.aimodeloverview.com/mcp",
      "headers": { "Authorization": "Bearer llmo_sk_..." }
    }
  }
}

Tool	What it answers
check_deprecation	Is this model deprecated or retired, and what replaces it?
list_models	Search and filter the catalogue.
get_model	The full record for one model.
list_providers	Discover valid provider slugs.
list_recent_changes	What changed in the dataset recently.

Limits, and what we record

The free tier allows 60 requests per minute, 5,000 per day and 50,000 per month. Quota is counted per organization across both protocols — extra keys do not buy extra calls, they exist so you can separate integrations and revoke them individually. Over the limit you get HTTP 429 with RateLimit-* and Retry-After headers; call /v1/me to see where you stand before that happens.

We record which endpoint or tool was called, when, with which key, whether it succeeded and how long it took. We do not record your query arguments, our responses, or your IP address. Raw events are deleted after 90 days. You can see your own usage in the dashboard.

Versioning

The version lives in the path. Fields get added to /v1; nothing is renamed or removed there. A breaking change would ship as /v2 alongside it.