Openclaw Cost Tracker Mcp

Tracks LLM token costs per agent and predicts rate limit exhaustion before you hit 429s. Reads JSONL cost logs with request-level provider, model, agent_id, token counts, and USD totals, then exposes tools for cost overview, top spenders, anomaly detection (per-agent median thresholds), and model routing recommendations with savings estimates. The v1.1 addition is quota window awareness: it projects which rate limit dimension will exhaust first based on your current burn rate versus provider headers, then suggests safe throttle targets. Built for OpenClaw native integration but works with any deployment that can wrap provider calls to emit the standard log schema. Cross-provider support includes Anthropic, OpenAI, Gemini, Ollama, and AWS Bedrock.