The Khiip server lets you capture, store, and semantically recall web content locally using a structured personal vault, with MCP integration for AI agents like Claude.

• Capture URLs (capture_url): Dispatch URLs to typed extractors for X/Twitter, Reddit, web articles, Wikipedia, YouTube, and PDFs (experimental). Stores raw source bytes plus a typed payload; idempotent by default (deduplicates by URL hash), with an option to force a new capture.

• Semantic Recall (recall): Search captured content using natural-language queries, ranked by cosine similarity using bundled MiniLM-L6 ONNX embeddings — no LLM cost, works fully offline after initial model download.

• List Captures (list_captures): Browse vault contents newest-first, with optional filtering by source platform (x, reddit, web, wikipedia, youtube, pdf) and pagination.

• Get a Capture (get_capture): Retrieve a single capture by ID in multiple formats — JSON payload, vault Markdown, YAML frontmatter only, or legacy Markdown.

• Refetch a Capture (refetch_capture): Re-process an existing capture along five independent dimensions:

  • extraction — network re-fetch to create a new superseding capture

  • re-extract — offline re-derivation of typed payload from preserved source bytes

  • re-render — offline re-rendering of Markdown body from existing typed payload

  • media — re-walk the media fetcher registry to update download statuses

  • wayback — re-submit the URL to the Wayback Machine (Save Page Now)

• Daemon Status (daemon_status): Check daemon health and metadata — confirms reachability and reports available extractors, embedder model/dimensions, and Wayback liveness.

Enables capturing Reddit posts and top-level comments, preserving crosspost lineage and removed-status information.

Enables capturing Wikipedia article content along with references and infobox data.

Enables capturing YouTube video metadata and transcripts.