Skip to content

Integrate DataHub MCP context with mem9 recall #340

Description

@King-Dylan

Purpose

Introduce DataHub MCP as a read-only external context provider for mem9 recall.

When a user asks about datasets, dashboards, metrics, schema, owners, quality, freshness, or lineage, mem9 should continue returning its own long-term memories as usual and optionally attach DataHub metadata context such as entity details and one-hop upstream/downstream lineage.

This keeps the product boundary clear:

  • mem9 remains the system of record for user/agent memory, conversations, preferences, and historical observations.
  • DataHub remains the system of record for enterprise data assets, metadata, ownership, lineage, and quality signals.
  • The integration lets agents combine both contexts at recall time without copying DataHub into mem9 or writing mem9 data back to DataHub yet.

Initial Scope

  • Add a configurable DataHub MCP external context provider.
  • Query DataHub MCP for data-asset-shaped recall requests.
  • Orchestrate search, get_entities, and get_lineage.
  • Return DataHub context alongside normal mem9 memories through external_context.
  • Keep DataHub retrieval read-only and gracefully degrade when DataHub enrichment or lineage fails.

Next Step

Add an optional mem9-to-DataHub publish path for curated data-asset observations.

mem9 can classify and reconcile memories that are clearly about DataHub entities, map them to DataHub URNs, then publish selected metadata-grade signals back to DataHub, for example:

  • dashboard/data issue observations
  • freshness or quality annotations
  • owner or team hints
  • business glossary notes
  • incident/context notes linked to datasets, dashboards, or metrics
  • custom DataHub aspects for agent-generated observations

The write path should be selective, permissioned, and auditable. Raw personal memory, private conversation history, and unrelated agent memory should stay in mem9.

Product Direction

This makes the integration complementary rather than overlapping: mem9 handles long-term user/agent context, while DataHub handles governed enterprise data-asset graph context.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions