DiscoMCP has 9 tools with complete input schemas and descriptions present for all tools and parameters. Naming follows verb_noun convention (inspect_target, plan_probe, execute_probe, profile_target, lookup_skill, refresh_profile, regenerate_skill, list_widgets, update_widget). However, descriptions are terse (10-50 chars for most tools), lacking WHEN/WHY context and recovery guidance. Output schemas are not documented in the source. Error handling is not visible in tool definitions. Tool composition is reasonable but some tools conflate multiple concerns (profile_target does discovery + profiling + artifact generation). Parameters lack detailed format/constraint documentation beyond enums.

• Expand tool descriptions to 100-200 chars. For each tool, add: (1) what it does, (2) when to call it instead of similar tools, (3) what it returns. Example: 'Inspect cached target declarations without executing. Call this first to see available tools and resources before planning a probe. Returns tool count, resource count, and risk classification.'
• Document output schemas for all tools. Add a 'Returns' section to each tool description listing key fields (e.g., 'Returns: {target_id, server_name, tools: int, resources: int, tool_cards: [{name, risk}]}'). This enables LLMs to chain tools correctly.
• Split profile_target into three tools: discover_target (static discovery only), profile_target (bounded profiling), and generate_artifacts (artifact generation). This lets agents compose them as needed and retry individual steps.
• Add format constraints to parameters. For 'output' (path), add 'Must be a valid directory path; will be created if missing.' For 'goal' (string), add 'Optional profiling objective (1-200 chars); if omitted, defaults to full discovery.'
• Add error handling guidance to tool descriptions. Example: 'On target not found, returns error with available targets. On safety violation, returns error with constraint details and suggests dry_run=true. Retryable errors include network timeouts.'

Inferred effective spec: 2025-06-18+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry