24 tools with complete input schemas and descriptions present. Naming follows verb_noun convention (exec, upload, download, mcp-add, etc.). Descriptions are concise (30-150 chars typical) and action-oriented. However, output schemas are not documented, LLMs cannot predict response structure for chaining. Parameter descriptions are minimal (10-30 chars) and lack format/constraint details. No error handling guidance. Risk annotations (WRITE/READ_ONLY) are present but not integrated into tool definitions. Missing idempotency, retry, and recovery patterns.

• Document output schemas for all tools. E.g., mcp-ls should specify: returns {servers: [{name: string, type: 'stdio'|'http'|'sse', status: 'connected'|'error'}], total: number}.
• Expand parameter descriptions to 50-100 chars with format/constraint details. E.g., 'ttl_secs: Time-to-live in seconds (default 900, range 60-86400)' instead of 'Invitation time-to-live in seconds'.
• Add error handling guidance to tool descriptions. E.g., 'exec: Returns exit code and stderr. Retryable on timeout; non-retryable on permission denied.'
• Implement confirmation pattern for destructive tools. Add optional 'dry_run' boolean param to upload, mcp-rm, edge-rm, control-sync. Return preview of changes before execution.
• Add discovery tools: list_edges() and describe_edge(name) to help agents discover available edges and their capabilities before invoking exec, upload, etc.