Two well-named, read-only tools with clear descriptions and proper input schemas. search_corpus and get_file follow verb_noun naming and include parameter types/descriptions. However, output schemas are not formally documented in the code, only described in docstrings. Error handling returns structured dicts but lacks recovery guidance (e.g., 'try search_corpus to discover paths'). Parameter constraints (k: 1-20, query: min 3 chars) are enforced at runtime but not declared in schema. Tool annotations (readOnlyHint) are present, which is good. Missing: explicit output schema documentation, per-parameter validation hints in descriptions, and actionable error messages that guide the LLM to next steps.

• Add JSON Schema output definitions to both tools. Document search_corpus returns {query: string, results: [{score: float, path: string, name: string, heading_path: string|null, text: string}]} and get_file returns {path: string, chars: int, text: string} or {path: string, source: string, text: string} or {error: string, hint?: string}.
• Declare parameter constraints in the input schema: search_corpus.k with minimum: 1, maximum: 20; search_corpus.query with minLength: 3. Remove runtime validation duplication.
• Enhance error messages with actionable recovery hints. E.g., 'path not in the indexed corpus, use search_corpus("<topic>") to discover valid paths' instead of relying on a separate hint field.
• Document result truncation limits in tool descriptions: 'Returns up to k chunks (text truncated to 2000 chars per chunk)' and 'Returns full indexed document (text truncated to 50000 chars)'.
• Unify get_file response structure: always return {path, text, source?, chars?, error?} so LLMs see a consistent schema regardless of success/failure path.