MCPMender exposes 4 tools via STDIO with minimal schema documentation. Tool definitions lack explicit input/output schemas in the source code provided. The 'probe' tool has a partial schema visible (timeoutMs, serverNames), but 'scan', 'repair', and 'rollback' show no input schemas in the provided code. Descriptions are present but generic (10-60 chars), below the 50-200 char baseline for LLM-optimized docs. No parameter descriptions visible for most tools. No error handling guidance, output schemas, or recovery patterns documented. Tool names are action-oriented (scan, probe, repair, rollback) but lack clarity on what distinguishes repair from rollback or what 'safe repairs' entails.

• Add explicit JSON Schema input definitions for all 4 tools. For 'repair', document the structure of repair objects (e.g., {type: string, target: string, action: string}). For 'rollback', clarify entryId format and how to discover valid IDs.
• Expand tool descriptions to 80-150 chars. Example: 'scan' → 'Scan all configured MCP clients and servers for configuration errors, missing dependencies, and connection issues. Returns a detailed report with severity levels and repair suggestions.'
• Add parameter descriptions. For 'probe': 'timeoutMs: Connection timeout in milliseconds (1000-30000, default 5000). serverNames: Optional array of specific server names to test; omit to probe all configured servers.'
• Document output schemas. For 'scan', specify: {clients: [{displayName, configFound, findings: [{severity, titleKey, detailParams}]}], summary: {detectedClients, configuredServers, errors, warnings, safeRepairs}}.
• Add error handling guidance. For 'repair': 'If repair fails, check the error message for the specific configuration issue. Use scan() to re-diagnose. If a repair is unsafe, it will be rejected with details.' For 'rollback': 'If entryId not found, call scan() to list available backup history.'