Single tool 'scan_mcp_server' has a reasonable name and description (89 chars), but critical gaps undermine quality. Input schema is present with two parameters (source, filePath), both with descriptions. However, the output schema is completely undocumented, no return type specification visible in the source. The tool description lacks actionable guidance on what the scan returns, when to use it, or how to interpret results. No error handling guidance provided. The tool appears to be a static analysis scanner, but the definition does not explain what vulnerabilities it detects, what the output structure is, or how an LLM should act on results. This is a read-only tool (no state modification), which is correct, but the definition quality is insufficient for reliable LLM integration.

• Document the complete output schema: return type (object), all fields (e.g., vulnerabilities[], severity, rule_id, line_number, remediation), and field types. Example: {vulnerabilities: [{rule: string, severity: 'critical'|'high'|'medium'|'low', line: number, message: string}], summary: {total: number, critical: number}}
• Expand tool description to 150-250 chars explaining: (1) what it does (scans MCP server source for security vulnerabilities), (2) when to use it (before deploying an MCP server), (3) what it returns (list of findings with severity and remediation). Current 89 chars is too sparse.
• Add parameter constraints: 'source' should specify max length (e.g., 1MB), supported encodings (UTF-8), and what happens if code is too large. 'filePath' should clarify accepted formats (relative path, filename, or full path) and supported languages (TypeScript, Python, JavaScript).
• Add error handling guidance: 'If source code is invalid or unsupported language, returns error with language detected and supported languages list. If source exceeds 1MB, returns error with truncation guidance.'
• Document the 13 OWASP MCP rules in the tool description or as a separate discovery tool. LLM needs to know what vulnerabilities are checked (command injection, path traversal, missing auth, secrets exposure, prompt injection, etc.) to understand results.