mcp-wallfacer is a TESTING HARNESS, not a production MCP server. Its 30 tools are intentionally buggy examples designed to validate fuzzing and invariant-testing capabilities. Most tools have minimal descriptions (many under 20 chars, e.g., 'BUG: terminates the process on any call'), no parameter descriptions, and schemas that are present but trivial (empty objects or single fields). Tools like 'crashes_now', 'hangs_forever', 'bad_protocol', 'run_shell', 'query_db' are deliberately unsafe. This is NOT a tool suite for agent use, it's a test corpus. Scoring reflects that: naming is clear (verb-based), but descriptions are diagnostic rather than LLM-optimized, and parameter documentation is absent. The 'sane' tools (echo, add, counter_get, slow_op) have slightly better structure but still lack rich descriptions.

• Expand all descriptions to 50 - 200 chars, explaining WHAT the tool does, WHEN to use it, and what it returns. E.g., 'crashes_now' → 'Test harness tool: intentionally terminates the process on any call. Used to validate crash detection in fuzzing harness. DO NOT use in production.'
• Add parameter descriptions for all inputs. E.g., 'limit' in paginate → 'Maximum items to return (1 - 50). Pagination test: tool occasionally returns one extra item.'
• Document output schemas explicitly. For record_create, specify: 'Returns {id: integer, event: string, created_at: ISO8601}' so agents know what fields to expect.
• Add error recovery guidance. E.g., divide → 'Error: division by zero. Ensure b > 0 before calling.' This helps test harnesses validate error handling.
• Clarify that this is a TEST HARNESS, not a production tool suite. Add a server-level warning in the capabilities response or README that these tools are intentionally buggy for validation purposes.

Inferred effective spec: 2025-06-18+.

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