Find the cache tax draining your AI bill: a cross-provider cache-TTL-waste detector, model-swap savings simulator, pricing-drift monitor, and per-agent cost attribution tool. MCP server + CLI.
How to pay
Subscribe
$19/month
Predictable monthly cost with included usage. Best for steady, high-volume traffic.
- Unlimited tools within plan limits
- One API key, billed once a month
- Cancel any time
Overview
Scorchmark is an observability and FinOps MCP server designed to catch the "invisible" costs in AI agent workflows that standard provider dashboards ignore. By ingesting your raw request logs—including Claude Code transcripts—it performs granular analysis on cache utilization, model economics, and runaway agent loops. Unlike passive billing portals, Scorchmark proactively identifies the specific mechanics causing budget spikes before they hit your credit card.
Key Capabilities
- Intelligent Waste Detection: Use detect_cache_waste to expose "cache-TTL tax" where loop intervals exceed your cache lifespan, causing expensive re-writes.
- Predictive Guardrails: Deploy check_budget and detect_spend_acceleration to catch exponential context growth or runaway loops in real-time, long before a billing limit is reached.
- Economic Simulation: Run simulate_model_swap for precise, per-row cost projections across different providers, and use detect_pricing_drift to alert you when silent provider rate changes threaten your unit economics.
- Granular Attribution: Leverage cost_by_agent to break down spend by specific agent IDs, providing the visibility needed to identify which autonomous processes are driving your infrastructure costs.
Use Cases
- Debugging Agent Loops: Use detect_stuck_agent and find_spend_anomalies to identify agents that are caught in recursive tool-calling loops or processing outlier-heavy requests that inflate costs.
- Proactive Rate-Limit Management: Monitor predict_rate_limit to receive status alerts when your token or request consumption suggests you are approaching exhaustion, allowing for dynamic throttling.
- Automated Alerting: Integrate build_alert_payload with your preferred notification stack (Slack, PagerDuty, or webhooks) to trigger immediate warnings when an agent's check_budget status shifts from "ok" to "breach."
Who This Is For
Scorchmark is built for AI engineers, FinOps practitioners, and developers managing production-grade autonomous agents. It is optimized for teams operating at scale who need to move beyond aggregate cloud billing and require microscopic, actionable insights into their model consumption, cache efficiency, and provider-specific pricing volatility.