The local MCP server exposes timelines, activities, and tag
writes but no way to reach captured screenshots. This blocks
on-device AI pipelines that would enrich time tracking with
vision/OCR — auto-tag suggestions from on-screen content,
searchable visual history, session recaps.

Proposal — two tools, paths only, reusing the existing query
surface:

• listscreenshots(fromTime, toTime, filter?)
  Returns [{ id, timestamp, absolutePath, width, height,
  sizeBytes, linkedActivityIds }]. Filter reuses the OData
  grammar from getcombined_activities so queries like
  "screenshots while Tag contains 'CW-10269'" work without
  new syntax.

• getscreenshotpath(id)
  Returns the absolute path for a single screenshot.
  Convenience wrapper for follow-up lookups.

Binary content stays on disk. The consumer opens files
directly. Zero MCP bandwidth, no base64 in context.

Ideal integration: add Screenshot as a new SummaryType that
must be explicitly requested in getcombinedactivities —
include "Screenshot" in the summaryTypes array to have it
returned. Existing callers see no change in behaviour;
screenshot data only enters results when the caller opts in.