🚀 Release: mcp-audit v0.7.0

[nathanschram]

🎉 mcp-audit v0.7.0 is here!

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

---

🆕 What's New in v0.7.0

Added

TUI Session Browser (mcp-audit ui)

• Interactive Session Browser — New mcp-audit ui command for exploring past sessions (#27)
  • Browse sessions across all platforms with list and detail views
  • Platform filter cycling with f key (all → claude-code → codex-cli → gemini-cli)
  • Date range filtering and search functionality
  • Drill-down into individual sessions with full token breakdown
• Session Pinning — Mark important sessions for quick access (#28)
  • p key to pin/unpin sessions
  • Pinned sessions persist across browser restarts
  • Preferences stored in ~/.mcp-audit/preferences.json
• Accuracy Display Indicators — Visual accuracy labels in session browser (#29)
  • Shows [exact], [estimated], or [calls-only] per session
  • Helps identify which sessions have native vs estimated token counts
• TUI Keybindings & Navigation — Full keyboard navigation (#31)
  • j/k or arrows: Navigate sessions
  • Enter: View session details
  • s: Cycle sort order (date, cost, duration, tools)
  • r: Refresh session list
  • q: Quit browser
  • ?: Show help overlay
• Tool Detail View — Drill-down into individual tool usage (#58)
  • View per-tool token consumption and call counts
  • Identify which tools consume most tokens

Live TUI Enhancements

• Smells Panel — Real-time efficiency issue detection in live TUI (#30)
  • Shows detected smells (HIGH_VARIANCE, TOP_CONSUMER, etc.) as they occur
  • Updates dynamically during session tracking
• Static Cost & Zombie Tools in Header — Schema overhead visibility (#60)
  • Header shows static cost when available
  • Zombie tool count displayed for quick awareness
• Reasoning Tokens Display — Thinking token tracking in TUI (#63)
  • Shows reasoning tokens for Gemini CLI and Codex CLI (o-series models)
  • Auto-hides for Claude Code (no thinking tokens exposed)
• Session ID & Files Monitored — Enhanced header information (#64)
  • Shows current session ID in header
  • Displays count of files being monitored
• Rate Metrics — Real-time tokens/min and calls/min (#65)
  • Automatically calculates based on session duration
  • Formats large rates (K/min, M/min) for readability
• Cache Hit Ratio — Token-based cache utilization metric (#66)
  • Shows percentage of input coming from cache
  • Distinct from cost-based efficiency (which factors in pricing)
• Unique Tools Count — MCP Servers panel shows tool diversity (#66)
  • Format: MCP Servers (4 servers, 12 tools, 47 calls)
  • Helps identify over-reliance on specific tools

Other

• AI Export Enhancements — Rate metrics and cache hit ratio included in exports
  • Both mcp-audit export ai-prompt and live TUI export updated
  • Provides richer context for AI analysis
• Cross-platform Keyboard Input — New keyboard.py module for consistent key handling
  • Works across macOS, Linux, and Windows terminals
  • Handles special keys (arrows, enter, escape, etc.)
• Preferences System — User preference persistence
  • New preferences.py module for storing user settings
  • Currently used for session pinning, extensible for future preferences

Changed

• Token Usage panel — Increased from 8 to 9 rows to accommodate rate metrics
• MCP Servers panel title — Now includes server count, unique tools, and total calls
• Project name detection — Shows full worktree path (e.g., project-name/main instead of just main)

Fixed

• models_used empty for single-model sessions — Now populated with session model even without MCP calls (#55)

Notes

• Schema version remains v1.6.0 (no data structure changes)
• Backward compatible: All changes are display-only enhancements
• New files: display/session_browser.py (1259 lines), display/keyboard.py, preferences.py

➡️ View Milestone

➡️ View all v0.7.0 issues

---

⚡ Install or Upgrade

# New install
pipx install mcp-audit

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

---

🔗 Resources

📦 PyPI	pypi.org/project/mcp-audit/0.7.0/
📋 Changelog	Full version history
📚 Docs	README & platform guides
🗺️ Roadmap	What's coming next

---

🤖 Supported Platforms

Platform	Token Accuracy
Claude Code	Native (100%)
Codex CLI	Estimated (99%+)
Gemini CLI	Estimated (100% with tokenizer)

---

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

Made with 🐻 by Little Bear Apps