Budget-Aware Code Memory

Token-budget-aware code indexing MCP server that retrieves relevant code by walking dependency graphs within specified token limits for 155 programming languages.

This server provides token-budgeted code retrieval by walking outward through code dependency graphs and returning the most structurally relevant code until the token budget is exhausted. It supports 155 languages via CodeGraphContext, delivers 0.07ms average query latency, and includes 22 tools for indexing, retrieval, discovery, and context analysis. Available as the npm package budget-aware-mcp.