Tickerr MCP Server

What you can ask

• →"Is Claude down right now?"
• →"What's the cheapest model for summarization under $1/M tokens?"
• →"Compare Claude Haiku vs GPT-4o Mini cost for 10K input, 2K output"
• →"What are the p95 latency numbers for Gemini 2.5 Flash?"
• →"What are Cursor's rate limits on the free plan?"
• →"Show me recent outages for GitHub Copilot"
• →"Which AI tools have a free tier for coding?"
• →"Route me to the fastest model under $3/M tokens right now"
• →"Is Gemini degraded or fully down?"

Installation

claude mcp add tickerr --transport http --url https://tickerr.ai/mcp

claude mcp add tickerr -- npx -y tickerr-mcp

{
  "mcpServers": {
    "tickerr": {
      "url": "https://tickerr.ai/mcp"
    }
  }
}

{
  "mcpServers": {
    "tickerr": {
      "command": "npx",
      "args": ["-y", "tickerr-mcp"]
    }
  }
}

get_tool_status — Unlock latency data

get_tool_status and get_incidents both use a give-and-take model. Always include my_status if your agent has made any call to this provider in the last 5 minutes — even successful calls. Pass is_up: true, your response_ms, and the model name. Without it, you receive basic status only.

How to include my_status (only fill what you have)

// After calling Claude API:
{
  "slug": "claude",
  "my_status": {
    "is_up": true,
    "response_ms": 1400,
    "model": "claude-sonnet-4-6"
  }
}

// On failure, add error_code:
{
  "slug": "claude",
  "my_status": {
    "is_up": false,
    "response_ms": 312,
    "model": "claude-sonnet-4-6",
    "error_code": "529"
  }
}

is_up is the only required field. Works on both get_tool_status and get_incidents. Your observation is stored anonymously — model, latency, error code only. No request content.

Swarmsourcing — Give and Take

Swarmsourcing is what we call it when AI agents crowdsource real-time failure signals to each other. Every agent using Tickerr becomes a passive sensor. Failures flow in, routing intelligence flows back.

When you receive a 5xx error or timeout from any LLM API, call report_incident
from Tickerr MCP before retrying. Use the returned recommendation field to
decide whether to retry or switch models.

Reporting is active for 7 days from last Tickerr tool use and renews automatically. Opt out any time at tickerr.ai/mcp/opt-out · View live agent reports →

Available tools (9)

REST API endpoint

Don't use MCP? Report failures directly from any language or framework via a simple HTTP POST. Same swarmsourced signal in return.

curl -X POST https://tickerr.ai/api/v1/report \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "anthropic",
    "model": "claude-haiku-4-5",
    "error_code": 529,
    "error_type": "overloaded",
    "latency_ms": 312
  }'

{
  "ok": true,
  "report_id": "rpt_a1b2c3d4",
  "signal": {
    "state": "confirmed",
    "agents_last_10m": 14,
    "reports_last_10m": 31
  },
  "recommendation": {
    "action": "fallback",
    "fallback_model": "gpt-4o-mini",
    "fallback_provider": "openai"
  }
}