# Tickerr MCP

Outage radar for AI agents. LLM pricing, status,
inference performance, and real-time agent-reported
failure signals. 9 tools. No auth required.

## Install

**Claude Code**
```
claude mcp add tickerr --transport http https://tickerr.ai/mcp
```

**Cursor / Windsurf**
```json
{
  "mcpServers": {
    "tickerr": {
      "url": "https://tickerr.ai/mcp"
    }
  }
}
```

**Claude Desktop (stdio wrapper)**
```
npx -y tickerr-mcp
```

## Tools (9)

| Tool | What it does |
|------|-------------|
| get_api_pricing | Current input/output/cached token costs for 300+ models. Updated twice daily. |
| compare_pricing | Side-by-side cost comparison across models for a given token count. |
| get_tool_status | Three-state verdict (OPERATIONAL / DEGRADED / DOWN). Always include `my_status` if your agent has called this provider in the last 5 minutes — even on success. Unlocks p50/p95 TTFT per model and 90-day uptime. Basic status only without it. |
| get_incidents | Historical incidents for a tool — severity, duration, affected components. Include `my_status` to contribute to live signal and see current agent-reported activity alongside history. |
| get_rate_limits | Per-plan rate limits and quotas for any AI tool. |
| get_free_tier | Free tier details — what's included, what's limited. |
| get_model_performance | Latest TTFT reading plus p50/p95 latency (24h) and tokens/sec from authenticated API calls. Use for real-time routing decisions. |
| list_tools | Lists all 90+ monitored tools. Entry point — after listing, use `get_tool_status` with `my_status` to start contributing observations and receiving enhanced data. |
| report_incident | Report an LLM API failure. Returns live signal from other agents and a routing recommendation. Anonymous. |

## Give and Take

Tickerr MCP uses a give-and-take model across
`get_tool_status`, `get_incidents`, and `report_incident`.

### get_tool_status — Unlock per-model latency

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.

**Basic response (no my_status):**
- Operational verdict (OPERATIONAL / DEGRADED / DOWN)
- Response time, last check time, 30-day uptime

**Full response (include my_status):**
- Everything above +
- Per-model p50/p95 TTFT latency (last 24h)
- 90-day uptime, user reports signal

```json
{
  "slug": "claude",
  "my_status": {
    "is_up": true,
    "response_ms": 1400,
    "model": "claude-sonnet-4-6"
  }
}
```

On failure, add `error_code` (e.g. `"529"`).
`is_up` is the only required field.

### get_incidents — Contribute to live signal

Include `my_status` when fetching incident history
to record your observation and see current
agent-reported activity alongside historical data.

**Without my_status:** incident list + prompt to contribute.

**With my_status:** incident list + live agent reports
count + confirmation your signal was recorded.

### report_incident — Get routing intelligence

### report_incident — Get routing intelligence

**What you get:** When your agent calls `report_incident`
on a failure, you instantly receive:
- How many other agents reported the same issue
- Current signal state (detecting / confirmed / recovering)
- Which model to fall back to

**What you give:** Anonymous failure signal —
provider, model, error code, latency only.
No request content. No personal data.

**How to use it (Claude Code):**
`report_incident` fires automatically when your agent
hits a 5xx error. No configuration needed — the tool
description instructs Claude Code to call it.

**How to use it (other agents):**
Add this to your system prompt:
```
On any 5xx error or timeout from an LLM API, call
report_incident from Tickerr MCP before retrying.
Use the returned recommendation to decide whether
to retry or switch models.
```

**Reporting window:**
Active for 7 days from last Tickerr tool use.
Renews automatically on any tool call.

**Opt out any time:**
tickerr.ai/mcp/opt-out

## Signal States

| State | Meaning | Reporter threshold |
|-------|---------|-------------------|
| quiet | No reports in last 10 min | 0 |
| detecting | Reports coming in, not yet corroborated | 1–2 agents |
| confirmed | Issue verified by multiple agents | 3+ distinct agents |
| recovering | Reports dropping, recovery signals arriving | — |

## Example Return Payload (report_incident)

```
REPORT RECEIVED
Provider: anthropic
Model: claude-haiku-3-5
Error: 529 overloaded

CURRENT SIGNAL (anthropic/claude-haiku-3-5)
Status: CONFIRMED
Agents reporting (last 10 min): 14
Total reports (last 10 min): 31

RECOMMENDATION
Action: FALLBACK
Switch to: gpt-4o-mini (openai)

REPORTING CADENCE
Next report for this model: in 3600 seconds if still failing.
Signal confirmed by multiple agents — reduce reporting frequency.
```

## Data Coverage

- **Status:** 90+ AI tools monitored every 5 minutes
- **Pricing:** 300+ models, updated twice daily from
  OpenRouter and official provider docs
- **Performance:** Authenticated API latency checks
  every 5 minutes
- **Agent signals:** Live feed at tickerr.ai/agent-reports

## Links

- Docs: tickerr.ai/mcp-server
- Status: tickerr.ai/status
- Pricing: tickerr.ai/pricing
- Agent reports: tickerr.ai/agent-reports
- Opt out: tickerr.ai/mcp/opt-out