token-optimizer-mcp

Overview

This skill provides an MCP server and client integration designed to drastically reduce context window waste during AI-assisted development. By replacing raw file reads, searches, and edits with optimized tools like smart_read, smart_glob, and smart_edit, it prevents repetitive and expensive operations. Additionally, it maintains a local knowledge graph that preserves findings, decisions, and dead ends across sessions, eliminating the need for models to re-derive past conclusions.

Capabilities

• ▸Differential file reading and editing for large files
• ▸Path-only globbing and optimized searching
• ▸Per-project knowledge graph tracking findings, decisions, and dead ends
• ▸Token accounting and live dashboard reporting across multiple CLI agents
• ▸Zero-turn refusals providing cached answers directly

Best for

Minimizing token waste when reading or editing large codebases over multiple agent sessions, Preserving architectural decisions, findings, and dead ends between agent interactions without using cloud RAG, Tracking and auditing net token savings and context transport across multiple AI coding clients

Works with

Security considerations

The tool operates locally on the user's machine, storing project knowledge graphs and session ledgers locally without sending telemetry or data to external third-party servers.

Categories