D-Eye provides 10 tools with consistent naming (verb_noun pattern: search, fetch, extract, query_evidence, export_research_packet, semantic_search, repo_inspect) and reasonable descriptions (avg ~80 chars). All tools are READ_ONLY and well-scoped. However, parameter descriptions are minimal or absent in the schema definitions visible in mcp_server.py. Input schemas are present but lack detailed type constraints, enums, or validation guidance. Output schemas are not formally documented. Error handling is implicit (via exceptions) rather than explicit recovery guidance. No tool annotations (readOnlyHint, etc.) despite all being read-only.

• Add explicit output schema documentation to each tool. Use Pydantic models or TypedDict in the FastMCP decorator to make return types visible to the MCP client.
• Enhance parameter descriptions in FastMCP decorators. E.g., @server.tool() def search(query: str) -> dict: 'Search across registered connectors for a query. Returns top results with source URLs and snippets.'
• Add readOnlyHint=True to all @server.tool() decorators to signal safety and enable agent optimizations.
• Wrap error responses with actionable recovery guidance. E.g., if fetch() fails, return {error: 'Failed to fetch URL', suggestion: 'Try search() to find alternative sources'}.
• Document pagination and result limits for search() and semantic_search(). Add max_results parameter (default 20, max 100) and return total_count in response.