Loading ToolBench

Reading the current corpus…

Sovereign-MCP-Blueprints MCP server · D · ToolBench

Sovereign-MCP-Blueprints

github.com/WizardofTryout/Sovereign-MCP-Blueprints

GDPR-compliant, auditable MCP server for PostgreSQL with anonymisation layer, read-only enforcement, and immutable audit logging

<=2025-11-25STDIO1 tools

55/100

Definition Quality80

Protocol Readiness32

Supportability28

Spec posture

Inferred effective spec <=2025-11-25·current baseline 2026-07-28

Static source inference · medium confidence · detected: Logging

Deprecated protocol patterns detected

Summary

Single tool 'query_anonymized_customer_data' has well-structured JSON Schema with proper type definitions, constraints (minLength, maxLength, pattern, min/max), and clear descriptions for all parameters. Tool description is comprehensive (180+ chars) and explains the anonymization guarantee and row-limit enforcement. However, output schema is not documented in the source, the tool description does not specify what fields are returned or their types. Parameter descriptions are strong (ISO codes, segment enums, limit enforcement), but the tool lacks error handling guidance (what happens on invalid region? database timeout?). No tool annotations (readOnlyHint, idempotentHint) despite READ_ONLY risk classification. Naming is clear and verb-forward ('query_'). Overall solid for a single-tool server, but missing output documentation and error recovery patterns.

Tools (1)

query_anonymized_customer_dataread onlysource verified80/100

Query anonymised customer data filtered by region and optional segment. Returns customer records with PII stripped and row count limited by server-side enforcement.

Issues (5)

HIGHtool ↗

Output schema not documented. Tool description does not specify what fields are returned (e.g., customer_id, name, email, segment, region) or their types. LLMs cannot plan downstream operations or extract the right data without knowing the response structure.

query_anonymized_customer_data

HIGHrecovery-guide ↗

No error handling guidance. Tool description does not explain what happens on invalid region code, database timeout, or permission denial. LLMs receive raw errors with no recovery path (e.g., 'connection refused' vs 'invalid region code, must be ISO 3166-1 alpha-2').

query_anonymized_customer_data

MEDIUMtool-annotations ↗

No tool annotations despite READ_ONLY risk classification. Tool should declare readOnlyHint=true and idempotentHint=true in MCP schema to signal safety to agents. Current implementation relies on implicit trust.

query_anonymized_customer_data

MEDIUMconstrained-input ↗

Recommendations

Document the output schema in the tool description or as a separate schema field. Example: 'Returns an array of objects with fields: customer_id (string), anonymized_name (string), segment (string), region (string), created_at (ISO 8601 date).'
Add error handling guidance to the description: 'If region is invalid, returns error 'Invalid region code, must be ISO 3166-1 alpha-2 (e.g., DE, FR, NL).' If database is unavailable, returns 'Database connection failed, try again in 30 seconds.'
Add tool annotations to the MCP schema: readOnlyHint=true, idempotentHint=true to signal safety to agents.
Clarify segment=null behavior: 'Omit segment or set to null to return all customer segments (premium, standard, trial).'
Migrate from STDIO to Streamable HTTP transport to enable remote access and hosted client compatibility. Use uvicorn (already in requirements.txt) to expose an HTTP endpoint.

Spec posture evidence

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

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

Score history

Overall score trend

First recorded score · v2 rubric

55/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	55	<=2025-11-25	v2