Problem
When multiple agents collaborate (e.g., dynamo-audit writes findings, another agent consumes them), there's no efficient way to detect new or updated memories. The consuming agent must re-fetch everything and manually determine what's changed.
Proposal
Add a check_updates MCP tool:
check_updates(
agent_id: "dynamo-audit", // filter by source agent
tags: ["cost-optimization"], // optional tag filter
since: "2026-03-26T07:43:00Z", // last check timestamp
)
Returns:
- Count of new memories since
since
- Count of updated memories since
since
- List of new/updated memory IDs with titles
- Suggested next
since value for the next poll
This enables efficient polling without re-downloading all content. The consuming agent can then fetch only the new/changed memories by ID.
Stretch: Watermark tracking
Store a per-consumer watermark so agents don't need to track since themselves:
check_updates(
agent_id: "dynamo-audit",
consumer_id: "claude-opus", // tracks watermark automatically
)
Problem
When multiple agents collaborate (e.g., dynamo-audit writes findings, another agent consumes them), there's no efficient way to detect new or updated memories. The consuming agent must re-fetch everything and manually determine what's changed.
Proposal
Add a
check_updatesMCP tool:Returns:
sincesincesincevalue for the next pollThis enables efficient polling without re-downloading all content. The consuming agent can then fetch only the new/changed memories by ID.
Stretch: Watermark tracking
Store a per-consumer watermark so agents don't need to track
sincethemselves: