The server has 6 tools with complete input schemas and basic descriptions. Naming follows verb-noun convention (search_exploits, get_exploit_path, etc.). However, descriptions are brief and lack context about when to use each tool relative to others. Parameter descriptions are present but sparse, many lack format guidance or constraints. Output schemas are undocumented; LLMs cannot predict the structure of results. Error handling is minimal: the code returns raw stderr and stdout without actionable recovery guidance. The tool set is reasonably composed (single responsibility per tool), but descriptions do not mention idempotency, side effects, or prerequisites. Baseline: production tools average 194 chars in tool descriptions and 72 chars in parameter descriptions; this server averages ~80 chars for tool descriptions and ~50 chars for parameters.

• Add explicit output schemas to all tools. Example for search_exploits: returns {type: 'object', properties: {results: [{edb_id: string, title: string, path: string}], total_count: number}}. This enables agents to extract EDB-IDs and chain them into get_exploit_content.
• Expand tool descriptions to 150-250 characters. Current average ~80 chars is below production baseline (194 chars). Example: 'Search for exploits in Exploit-DB by keyword or vulnerability type. Returns EDB-ID, title, and local path for each match. Use this to discover exploits, then call get_exploit_content to view details.'
• Add per-tool error recovery guidance. Example: If searchsploit not found, return 'searchsploit command not found. Install exploitdb via: pip install exploitdb, then retry. Or call update_exploitdb to initialize the database.' instead of a bare error message.
• Annotate side-effect tools with destructiveHint or modifiesState in descriptions. Example for mirror_exploit: 'Copies exploit to destination directory. WRITE operation: may overwrite existing files. Idempotent: safe to retry.' For update_exploitdb: 'Downloads and updates local Exploit-DB mirror. WRITE operation: may take several minutes. Non-idempotent: re-running will re-download the entire database.'
• Document parameter formats and valid values. For search_cve, update description to: 'CVE identifier in format YYYY-NNNNN (e.g. 2021-44228) or CVE-YYYY-NNNNN (e.g. CVE-2021-44228). Both formats are accepted and will be normalized.'
• Add result-limit and pagination guidance. In search_exploits and search_cve descriptions, state: 'Returns up to 100 matches by default. For very large result sets, use exclude parameter to narrow scope. Results are formatted as text; call get_exploit_path or get_exploit_content to retrieve structured data.'