Skip to content

fix: improve MCP path tracing and lineage evidence guidance - #145

Open
itu-itis24-buyukhelvacigilm24 wants to merge 1 commit into
datahub-project:mainfrom
itu-itis24-buyukhelvacigilm24:patch-1
Open

fix: improve MCP path tracing and lineage evidence guidance#145
itu-itis24-buyukhelvacigilm24 wants to merge 1 commit into
datahub-project:mainfrom
itu-itis24-buyukhelvacigilm24:patch-1

Conversation

@itu-itis24-buyukhelvacigilm24

Copy link
Copy Markdown

Summary

  • Prefer the shipped get_lineage_paths_between MCP tool for exact asset or column path tracing when it is available.
  • Keep datahub lineage path as the fallback and label bounded BFS results honestly.
  • Add concise reporting guidance for lineage granularity, completeness, empty results, observed facts, and agent inference.

Why

The current lineage skill recommends the CLI first for specific path tracing and implies that MCP may not support it. The official DataHub MCP Server now advertises get_lineage_paths_between for exact lineage paths between assets or columns.

While exercising the MCP lineage tools in a real agent workflow, we also found that agents need explicit evidence boundaries: dataset-level edges do not establish column lineage, a capped result is not a complete graph, and zero observed edges do not prove that no dependency exists.

Scope

This is a one-file guidance change. It does not modify DataHub core, the MCP server, or the existing /datahub-audit routing.

Validation

  • GitHub Markdown preview renders the new guidance correctly.
  • Comparison contains one changed file: 27 additions and 2 deletions.
  • Guidance was checked against the official MCP Server tool inventory.
  • Repository CI checks are pending.

Related

Complements #84, which updates get_lineage parameters and column traversal. This PR covers exact path tracing and evidence-safe reporting semantics.

Prefer the shipped get_lineage_paths_between tool when available, retain the CLI fallback, and clarify granularity, completeness, empty-result, and observed-vs-inferred reporting boundaries.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant