#cost-tracking

A local-first, multi-provider cost meter for LLM usage, exposed as MCP tools. Captures every call into a local SQLite ledger and lets any coding agent query spend, compare providers, and get recommendations — no cloud, no account. First-class support for Chinese providers (Qwen, DeepSeek) alongside Anthropic and OpenAI.

Burnwatch detects every paid service in your project, tracks spend across 14 AI-native tools (Anthropic, OpenAI, Vercel, Supabase, Stripe, Scrapfly, Browserbase, and more), and exposes real-time cost data via MCP tools — so any LLM can factor budget into its recommendations. Four tools: get_spend_brief (full project spend overview), get_service_spend (single service deep dive with gotchas and alternatives), detect_paid_services (scan any project), list_registry_services (browse the registry). Zero dependencies. Works with Claude Code, Cursor, or any MCP-enabled client.

The first MCP-native eval and observability tool for AI agents. Any MCP-compatible agent discovers and uses Iris automatically — no SDK, no code changes. Log traces with hierarchical span trees, evaluate output quality with 12 built-in rules (PII detection, prompt injection, cost thresholds), and track what your agents are actually doing and costing you. Real-time dark-mode dashboard, OpenTelemetry-compatible span structure, self-hosted with SQLite. MIT licensed.