🚀 Release: mcp-audit v0.6.0 #57

nathanschram · 2025-12-12T06:27:24Z

nathanschram
Dec 12, 2025
Maintainer

Find + fix context bloat & high token usage in your MCP tools.

Multi-Model Per-Session Tracking — Track token usage when sessions switch between models (#23)
- models_used: Array of unique model identifiers used during session
- model_usage: Per-model breakdown with tokens, cost, and call counts
- Call.model: Track which model handled each tool call
- TUI displays multi-model breakdown when multiple models detected
Dynamic Pricing via LiteLLM — Auto-fetch current model pricing from LiteLLM API (#24)
- 2,000+ models supported via LiteLLM pricing database
- 24-hour local cache with automatic refresh
- Graceful fallback chain: API → TOML config → built-in defaults
- pricing_source field in data_quality block (api/toml/default)
- Configure via mcp-audit.toml: enable/disable API, set cache TTL, offline mode
Static Cost Foundation — Infrastructure for MCP schema overhead tracking (#25)
- StaticCost dataclass for per-server schema weight
- Schema documented in data-contract.md
- Full implementation requires MCP protocol work (deferred to future release)
Schema v1.6.0 — Multi-model intelligence fields (#26)
- session.models_used: List of models used
- session.model_usage: Per-model statistics
- tool_calls[].model: Model per call
- data_quality.pricing_source: Where pricing came from
- data_quality.pricing_freshness: Age of pricing data

Pricing configuration — New API-first approach with TOML fallback
- Default behavior: Fetch from LiteLLM API, cache 24h, fall back to TOML
- Offline mode: Set [pricing.api] enabled = false for TOML-only
Test coverage — 56 new tests for v0.6.0 features (723 total)

# New install
pipx install mcp-audit

# Upgrade existing
pipx upgrade mcp-audit
# or: pip install --upgrade mcp-audit

💬 Questions, feedback, or want support for another CLI platform?
Reply here or start a discussion!

Made with 🐻 by Little Bear Apps