Circus MCP has 6 tools with explicit schemas and descriptions visible in the source code. However, the server suffers from several critical definition quality issues that prevent it from scoring higher. Tool descriptions are extremely terse (7-20 characters), providing minimal context for LLM selection. Parameters are present with type definitions but lack comprehensive descriptions explaining constraints, formats, or expected ranges. Output schemas are completely undocumented, callers have no visibility into what fields are returned or their structure. Error handling is minimal, with no guidance on recovery strategies. The tools themselves are well-named with clear verb-noun patterns (add_process, start_process, stop_process, restart_process, list_processes, get_process_status), which is a strength. However, the extreme brevity of descriptions (e.g., 'Add a new process to Circus', 'Start a process', 'Stop a process') falls well below the 50-200 character guideline for production-grade tools and provides insufficient context for reliable LLM reasoning about when and why to call each tool.

• Expand tool descriptions to 80-150 characters. Example for start_process: 'Start a stopped process by name. Requires process to exist and be in stopped state. Idempotent: restarting an already-running process is a no-op. Returns process status and uptime.' This clarifies intent, prerequisites, idempotence, and output.
• Document output schemas as structured objects. For add_process, return {process_id: string, name: string, status: 'added', created_at: timestamp}. For list_processes, return {processes: [{name, status, pid, uptime_seconds}], total: int}. For get_process_status, return {name, status, pid, uptime_seconds, cpu_percent, memory_mb}.
• Add parameter descriptions and constraints to add_process: 'numprocesses: Number of parallel process instances (1-1000, default 1). Must be a positive integer.' 'working_dir: Absolute or relative path where the process runs. Defaults to current directory. Must be readable and writable.'
• Implement pagination for list_processes. Add optional parameters: limit (default 20, max 100), offset (default 0). Return {processes: [...], total_count: int, limit: int, offset: int, has_more: bool}.
• Add error handling with recovery guidance. Instead of 'Failed to connect to Circus daemon', return 'Cannot reach Circus daemon on localhost:5555. Check: (1) circus-web service is running, (2) network connectivity, (3) CIRCUS_ENDPOINT env var. Retry in 30s or check system logs.'
• Clarify idempotence in tool descriptions. add_process should state: 'If process with this name already exists, returns existing process_id and status=already_exists (no duplicate created).' start_process should state: 'If process is already running, operation succeeds silently (idempotent).'