embgrep has 4 tools with basic descriptions and visible input schemas. However, critical gaps limit production readiness: (1) descriptions are minimal (10-50 chars), well below the 50-200 char LLM-optimized baseline; (2) output schemas are undocumented, tools return JSON strings but the structure is not declared; (3) no error handling guidance, tools fail silently or return generic JSON; (4) parameters lack constraints (e.g., top_k unbounded, patterns free-form); (5) no idempotency or retry semantics documented. Tools are named clearly (verb_noun pattern) and schemas are present, preventing a lower score, but the implementation is minimal and would not pass production code review.

• Expand tool descriptions to 80-150 chars. Example: 'index_directory: Recursively index files in a directory using semantic embeddings. Call this first to build the search index. Supports glob patterns (*.py, *.md). Returns count of files indexed and chunks created.'
• Document output schemas explicitly. For semantic_search, declare: 'Returns array of {file_path: string, score: float 0-1, line_start: int, line_end: int, chunk_text: string (max 500 chars)}'.
• Add parameter constraints: top_k as integer 1-100 (default 5); patterns as comma-separated glob list; path_filter as SQL LIKE pattern with escaping note.
• Implement error recovery hints. Catch FileNotFoundError and return 'Directory not found: {path}. Verify the path exists and is readable.' Catch empty index and return 'No index found. Call index_directory() first to build the index.'
• Document idempotency: 'index_directory is idempotent, calling twice with the same path and patterns re-indexes unchanged files (skipped via hash comparison). update_index is also idempotent and safe to retry.'