Single tool 'search_corpus' has a reasonable name and description (73 chars), but the input schema is minimal. The query parameter lacks a description in the schema itself, only the tool description provides context. No output schema is documented, leaving LLMs unable to plan downstream operations or extract structured results. The tool is read-only (good), but lacks pagination, result limits, and error recovery guidance. No parameter constraints (min/max length, format) are visible. Overall, the tool is functional but underspecified for production agent use.

• Add a detailed output schema documenting the structure of search results (e.g., array of {chunk_id, source, title, text, relevance_score}). This enables LLMs to extract and chain results.
• Add a description to the 'query' parameter in the JSON Schema: 'The search query text (1-500 characters). Refine if initial results don't answer the question.'
• Implement pagination: add 'limit' (default 20, max 100) and 'offset' (default 0) parameters, and return a 'total_count' field so LLMs know when to fetch more results.
• Document error cases and recovery paths: 'If no results found, try a broader query or check corpus status via [future status tool].'
• Add input validation: enforce query length (1-500 chars), reject empty strings, and return actionable errors like 'Query too short (min 1 char)' or 'Query too long (max 500 chars).'