FluxGuardian MCP server — exposes OpenMetadata tools for Claude Desktop
FluxGuardian MCP server has 6 well-named, read-only tools with clear descriptions (avg 156 chars) and proper input schemas using Pydantic validation. Naming follows verb_noun convention (search_assets, get_column_lineage, find_owner). All tools have descriptions and input schemas with enums for constrained values (change_type, asset type). However, output schemas are not documented in the source, responses are inferred from backend API calls but not formally declared. Error handling is minimal: the _get/_post helpers return None on failure with only logging, no actionable recovery guidance for LLMs. Parameter descriptions are adequate but lack format/range details (e.g., 'column_fqn' examples show FQN format but no explicit pattern constraint). No tool annotations (readOnlyHint, idempotentHint) despite all being read-only operations.
Find who owns a data asset — table, dashboard, ML model, or pipeline. Returns the owner's email and Slack handle so you know who to ping.
Compute the blast radius of a schema change — which downstream assets would break, who owns them, and the overall severity. Use this before merging any database migration PR.
Get the full data lineage for a specific column — both upstream sources and downstream consumers (dashboards, ML models, pipelines). Also returns PII tags on the column.
Get the full column schema for a table — column names, data types, nullable flags, and PII tags. Use this to understand table structure before writing migrations or queries.
List every column across all tables that has a PII (Personally Identifiable Information) classification tag — Email, Phone, Financial, etc. Use this for compliance audits or before sharing data access.
Search for data assets (tables, dashboards, ML models, pipelines) in OpenMetadata by name or keyword. Optionally filter by asset type. Use this to discover what assets exist before diving into lineage or ownership.
Output schemas not documented. Tools return JSON from backend but LLMs cannot see what fields to expect. Forces LLMs to guess at response structure and risks failed downstream chaining.
Error handling returns None with silent logging; no actionable recovery guidance. When backend is unreachable or returns 404, LLM receives no signal about what to do next (retry, ask user, try alternative tool).
No tool annotations despite all tools being read-only. Missing readOnlyHint and idempotentHint annotations prevent clients from optimizing caching and retry logic.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | <=2025-11-25 | v2 |
Parameter descriptions lack format constraints. 'column_fqn' shows examples (users.email, foodieexpress.public.users.email) but no explicit pattern or format rule. LLMs may pass invalid FQN formats.
list_pii_columns accepts no parameters but description does not explain pagination or result limits. If dataset is large, response could exceed context window without guidance.