Report when semantic query results are truncated

[vara-bonthu]

Description

The server applies a default row LIMIT of 1000 when no limit is provided. If more than 1000 groups exist, the response silently returns the first 1000 without indicating that the result is incomplete.

This can cause an agent or application to present a partial result as a complete answer. It also contradicts the documentation stating that requests are refused rather than truncated.

Proposed solution

• Execute bounded queries with LIMIT requestedLimit + 1
• Remove the extra row before returning the response.
• Add limit and hasMore or truncated to the response.
• Include completeness metadata in the result cache entry.
• Update MCP instructions so agents disclose incomplete results.
• Correct the server documentation.

Example

{
  "rowCount": 1000,
  "limit": 1000,
  "hasMore": true
}

Implementation areas

• internal/serving/limits.go
• internal/serving/service.go
• internal/serving/mcp/mcp.go
• website/src/content/docs/architecture/server.mdx