Add context composition endpoint for LLM-ready entity summaries

[ravisuhag]

Context

AI agents have limited context windows and need pre-composed, information-dense summaries rather than raw API responses. Compass should offer a context composition endpoint that aggregates everything known about an entity into a single LLM-ready document.

Current State

GetEntityContext RPC and the MCP get_context tool exist and return entity + edges + related entities. This issue tracks the remaining gaps to reach the full vision.

Scope

Enhance the context composition to return a composed document combining:

• Schema with column descriptions and types (from entity properties)
• Immediate upstream and downstream lineage (from edges)
• Current owners and responsible teams (from owned_by edges)
• Last schema change and recent modifications (from temporal entity history via valid_from/valid_to)
• Freshness and quality signals (from entity properties written by ingesters)
• Related entities and documentation links (from edges)

Design Considerations

• Output format should be structured markdown optimized for LLM context windows
• Support configurable depth and detail levels:
  • summary — for broad exploration
  • detailed — for SQL generation (full column-level detail)
  • diagnostic — for issue investigation (lineage + change history)
• Should be available through both REST/gRPC and the MCP server

Why

Raw JSON payloads waste context window tokens and require the AI agent to parse and reason over structure. A composed context document gives the agent exactly what it needs in a format it can reason over efficiently.

References

• Compass Roadmap — Context Composition