mcp-optimization

Token optimization for MCP servers involves selecting appropriate servers per task, implementing caching strategies, and budgeting tokens by session type. The Skill maps server roles to Quick fix, Feature dev, Architecture, and Audit/review scenarios, providing per-session token plans and pre-fetch patterns to maximize efficiency.

Establish session-type budgets and server-role mappings before execution. Apply caching and memory patterns to reduce redundant token consumption, use sequential-thinking discipline to avoid wasteful reasoning, and implement pre-fetch patterns to align server selection with available token capacity.

Token packing and caching reduce consumption across repeating tasks and multi-server workflows. Use caching for stable data retrieved across sessions and packing to consolidate requests. These strategies are especially valuable in resource-constrained environments or when token budgets limit session length.

Avoid selecting servers without role mapping to task type, ignore session budget constraints, or fetch data repeatedly without caching. Skip cost-optimization checks and don't apply sequential-thinking discipline. These anti-patterns waste tokens and reduce session effectiveness across Quick fix through Audit/review workflows.

AST-Grep rules enable precise code selection and pattern matching, reducing the volume of code passed to MCP servers. This targeted extraction minimizes token consumption while maintaining context accuracy, particularly during Architecture and Audit/review scenarios where selective analysis is critical.