Today’s Update: Cleaner “ask/extract” tools + stash reindexing

Two small-but-foundational improvements landed in mcp-research-friend: one makes the tool surface easier to reason about, and the other makes the stash easier to maintain over time.

Tool surface: “extract” vs “ask”
Q&A is now a first-class operation instead of being bundled into extraction. There are dedicated tools for asking questions about:

• a URL (friendly_web_ask)
• a stashed document (stash_ask)

Web pages and PDFs are also handled through a single extraction path (friendly_web_extract, which auto-detects HTML vs PDF), and fetch naming is clarified (friendly_web_fetch).

Stash: easier navigation, searching, and upkeep
A few stash changes reduce friction when you’re actually trying to retrieve something:

• stash_extract is focused on reading content, and supports jumping in by line number (useful when you want to go straight to a search hit).
• stash_search has more predictable matching and better controls (context sizing, doc ID filters, match limits), plus safer invocation details.

A new reindex_stash tool makes it possible to re-run classification/summaries across existing stashed documents and update stored metadata. Classification for long documents is improved by sampling text from across the document rather than relying on just the opening.

Guardrails
Topic names produced by classification are validated to avoid unsafe paths, and URL handling is restricted to http/https with checks that consider the final destination after redirects.