token-optimization

Reduce token consumption by enabling code execution with MCP, loading tool definitions on demand from the filesystem, and applying smart context window management to preserve crucial history while lowering costs.

Progressive tool discovery minimizes token waste by loading only the needed tool definitions on demand from the filesystem, avoiding the token bloat caused by loading entire tool catalogs into the context window at once.

Yes, you can use MCP for code execution to generate code that directly calls tools, keeping Claude's context lean by executing operations externally rather than embedding verbose tool outputs inline.

Implement prompt and response caching strategies to lower repeated token costs, pairing them with cost monitoring and token-budget aware features to track and optimize overall API expenditure.

Manage context window compaction by applying smart summarization techniques that budget the available token space, preserving crucial conversation history while reducing the overall token load to fit within limits.

No specific dependencies are required to implement token optimization with MCP; the skill provides approaches for filesystem-based tool loading, on-demand usage, and token-budget aware caching natively.