AI DownDetector MCP Server

AI DownDetector tracks outages for ten AI providers (OpenAI, Anthropic, Google Gemini, Meta Llama, Mistral, xAI, Microsoft, Amazon, Zhipu AI, Moonshot) from their official status pages plus API health checks, polled every five minutes. This MCP server lets AI agents read that status directly.

Endpoint and protocol

One POST-only endpoint speaking MCP over Streamable HTTP, protocol revision 2026-07-28 (2025-11-25-era clients work via the legacy initialize handshake). No sign-up is offered here: keys are issued by the site admin. Send your key as Authorization: Bearer aid_… on every tools/list and tools/call request.

curl -s -X POST https://ai-downdetector.com/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -H 'MCP-Protocol-Version: 2026-07-28' \
  -H 'Authorization: Bearer aid_YOUR_KEY' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Getting an API key

Contact the site admin (keys are issued manually; there is no self-serve signup).
Send the key as Authorization: Bearer aid_YOUR_KEY on every tools request.
Revoke or rotate by asking the admin; revoked keys stop working on the next call.

Tools

list_providersread-only

All tracked AI providers with slug, name, tier, data source and official status page URL.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": { "name": "list_providers", "arguments": {} }
}

get_provider_statusread-only

Current status of one provider: overall status, per-model status, last-polled time, staleness note and the official status page.

{
  "jsonrpc": "2.0",
  "id": 2,
  "method": "tools/call",
  "params": {
    "name": "get_provider_status",
    "arguments": { "provider": "openai-downdetector" }
  }
}

get_status_historyread-only

Per-day worst status for one provider over the last N days (1-30, default 7), plus up to 10 recent incidents. One provider per call.

{
  "jsonrpc": "2.0",
  "id": 3,
  "method": "tools/call",
  "params": {
    "name": "get_status_history",
    "arguments": { "provider": "anthropic-downdetector", "days": 14 }
  }
}

submit_reportwrite

Report that a provider is having problems right now (like the website's report button). Write-only acknowledgement; one report per provider per address per 10 minutes; never triggers the site's public community flag on its own.

{
  "jsonrpc": "2.0",
  "id": 4,
  "method": "tools/call",
  "params": {
    "name": "submit_report",
    "arguments": { "provider": "openai-downdetector" }
  }
}

Limits

60 tool calls per hour and 1,000 per day per key.
submit_report: 20 per day per key, and one report per provider per address per 10 minutes.
20 requests per hour per IP for calls without a valid key.
429 responses include Retry-After.
At most 10 results per call (models listed, incidents); history windows are capped at 30 days. No argument can raise these caps.
Privacy: per-call usage is counted (counts and timestamps only). No query text, no transcripts, no raw IP addresses are stored.

Attribution

Every tool result ends with a source line naming this unofficial dashboard and the provider's official status page — the official status page is always the authority. Personal and agent use is free; bulk copying or reselling the history data is not allowed.

Connect your client

Claude (web, desktop, mobile)

Settings → Connectors → Add custom connector, URL https://ai-downdetector.com/mcp, under Request headers add Authorization: Bearer aid_YOUR_KEY. Source: Claude docs.

Cursor

Add to ~/.cursor/mcp.json (source: Cursor docs):

{
  "mcpServers": {
    "ai-downdetector": {
      "url": "https://ai-downdetector.com/mcp",
      "headers": { "Authorization": "Bearer aid_YOUR_KEY" }
    }
  }
}

Claude Code

claude mcp add --transport http ai-downdetector https://ai-downdetector.com/mcp --header "Authorization: Bearer aid_YOUR_KEY"

Source: Claude Code docs.

ChatGPT

Custom MCP apps require developer mode; as of October 2026, full MCP support (including write actions) is limited to Business, Enterprise and Edu workspaces and needs an admin to enable developer mode. Source: OpenAI docs.

The protocol surface (server/discover, initialize, ping) works without a key so clients can identify the server; every tool call needs one.