structured-output-repair · packages

A Zod schema–driven repair engine that applies six graduated strategies (prose extraction, truncation repair, type coercion, fuzzy key matching, extra field removal, and input analysis) to malformed LLM outputs, returning schema-conforming data or detailed failure metadata. Exports `repair`, `repairOutput`, `isValid`, and `analyzeInput` functions, and requires `zod` as a peer dependency.

An MCP server that exposes `structured.repair` and `structured.analyze` as tools, letting MCP-compatible clients (Claude Desktop, Cursor) repair malformed LLM structured outputs against a JSON Schema.