token-optimizer-mcp

Tech stack & purpose

Token Optimizer MCP is an open-source project that measures and reduces token consumption for AI coding agents through caching, diffing, and compression techniques. The project provides an MCP (Model Context Protocol) server with tools like `smart_read`, `smart_glob`, and `smart_edit` that cut context and token usage by 60–90%, alongside a live local knowledge graph shareable across 16 CLI clients including Claude Code, Codex, Gemini CLI, and Amp. The implementation includes PowerShell hooks for Claude Code integration, evaluation suites covering command recovery and cross-project transfer, and comprehensive testing infrastructure built on Node.js with token counting via tiktoken. The project's evidence tiers range from deterministic runtime checks (29/29 passing, 1,000,000 canonical events processed at 44,475 events/second) through multi-model handoff testing, though it maintains an "insufficient" release verdict pending powered effectiveness studies and production traffic validation.

Languages

Contributors

• Franklin Moormann420 commits
• 
  dependabot[bot]32 commits · 0 followers
• franklinic31 commits
• 
  github-actions[bot]20 commits · 0 followers
• Christophe1 commits
• Mikhail Nemerov1 commits
• Christopher Mehlin1 commits