This is a mock MCP server with critically deficient tool definitions. All 5 tools have empty input schemas ({"type":"object"} with no properties), no parameter descriptions, and minimal tool descriptions (10-50 chars). Tool #5 (exfiltrate_secrets) is undeclared in the signed catalogue, a security red flag. The server appears designed to demonstrate policy enforcement and audit logging, not production tool quality. No output schemas documented. No error handling guidance. This is a reference implementation for testing obsign's policy layer, not a usable agent tool.

• Add explicit properties to each tool's input schema. E.g., delete_production_db should declare parameters like 'database_name' (string, required), 'confirmation_token' (string, required), with descriptions and constraints.
• Expand tool descriptions to 50 - 200 characters. Include: what the tool does, when to use it, what it returns, and any prerequisites. E.g., 'Permanently deletes a production database. Requires a confirmation token. Returns {success: bool, deleted_at: ISO8601}. Use only after backup verification.'
• Document output schemas for all tools. Specify return types, required fields, and field descriptions. E.g., ticket_update returns {ticket_id, status, updated_at, assignee_id}.
• Add parameter descriptions to all inputs. E.g., 'database_name: The production database identifier (alphanumeric, 3 - 64 chars). Must match an active database in the registry.'
• Implement confirmation/dry-run for destructive tools. Add a 'dry_run' boolean parameter (default false) and a 'confirmation_token' parameter. Return a preview before executing irreversible operations.

Inferred effective spec: <=2025-11-25.

• Relies on Sampling (deprecated) - integrate directly with the LLM provider API