MCP is stuck on retrieval: the harder problem is what you put in
kbavandi · reddit · 2026-08-26
The author of Kurator argues that MCP tools focus too much on retrieval (vector search, chunking), but quality is capped by input data. Storing URLs forces models to re-fetch cluttered pages. The author proposes 'Capture time': saving pages with custom prompts to extract structured context while reading. Kurator exposes this library via MCP, turning folders into structured records instead of link dumps.
More from coding & agent
- Dev claims he shipped 15-25 PRs a day before LLMs, while most devs now ship less — KyeGomezB · 2026-08-26
- Weekend Project: Streaming Peloton Data & Building Custom Android App with Claude — vykthur · 2026-08-26
- A full day with Claude Design: needs the $200 tier, beats iterating natively in Codex — zeeg · 2026-08-26
- WSJ: Agent workflows demand ultra-low latency compute — rohanpaul_ai · 2026-08-26
- Mixedbread Releases Toast 1: Deep Search Agent with Context Management — yacineMTB · 2026-08-26
- Budget policies for overnight background agents: preventing infinite retry loops — Some-Beat5994 · 2026-08-26