Solves the "which CSV column maps to which schema field" problem using seven different scoring strategies: exact name match, known aliases, initialism detection, pattern recognition for emails/UUIDs/dates, statistical profiling, fuzzy string matching, and optional LLM scoring. Combines them with the Hungarian algorithm for optimal one-to-one assignment, calibrates confidence scores, and spits out mappings with explainable reasoning. Useful when you're ingesting messy customer exports, vendor CSVs, or legacy database dumps and need to programmatically align them to your canonical schema without writing brittle manual rules. The MCP exposes map operations over stdio or HTTP, handling DataFrames, CSVs, databases, or in-memory records. Python and TypeScript implementations share a golden test suite for bit-identical decisions. Originally built by Ben Severn, now maintained in the goldenmatch monorepo.
claude mcp add --transport stdio benseverndev-oss-infermap uvx infermap