media-download

Media Download

Download media by URL and (natively) durably store it in the knowledge graph.

When to use

• Download a video/media file by URL (YouTube, Rumble, and other yt-dlp sites).
• Persist the raw bytes + yt-dlp metadata into epistemic-graph as a :MediaAsset (happens automatically when a live engine is reachable).

When NOT to use

• Audio-only extraction to MP3 → media-audio.
• Reading media already in the KG → query the :MediaAsset nodes directly.

Prerequisites & environment

Connect via the mcp-client skill against the media-downloader MCP server.

Variable	Required	Notes
GRAPH_SERVICE_ENDPOINTS	optional	Engine endpoint for native KG ingestion (e.g. tcp://host:9100); ingestion auto-no-ops if unset/unreachable

MCP_TOOL_MODE (condensed|verbose|both) selects the tool surface.

Tools & actions

Tool	Purpose
download_media	Download a URL; returns {status, file, kg_asset?}

Key parameters

• video_url — the media URL (required).
• download_directory — where to save the file (default .).
• audio_only — extract audio to MP3 (prefer media-audio for that intent).

Recipes

Download a video and store it in the KG:

{"video_url":"https://www.youtube.com/watch?v=<id>","download_directory":"/data/media"}

The response includes kg_asset ({asset_id, digest, size_bytes, media_type}) when the engine ingested the raw bytes.

Gotchas

• Native KG ingestion is default-on but best-effort: no engine reachable → the file still downloads, kg_asset is simply absent.
• Blobs are content-addressed + deduped: re-downloading identical bytes stores no new chunks.
• Large files stream into the blob store; peak memory is bounded, not file-sized.

Related

• Reuses the shared media ontology (:MediaAsset, :Blob, :DownloadJob) federated by the jellyfin-mcp package — media-downloader does not define its own media classes.
• Composed by: a media-librarian workflow that downloads then reasons over the KG.