Skip to main content
Keep Kino Desktop open and signed in on the agent’s machine. Connect the agent to http://127.0.0.1:45680/mcp using the setup instructions, then call getKinoInstructions. No infrastructure keys or browser cookies are needed.

​
Discover and read

Start with browseMoods({resource: "sources"}). Each entry retains its complete source record, including counts, facets, and any catalog coverage limitations. Use the returned slug:
Other resources are collections, channels, people, and styles. Their singular forms take a key. arena enumerates the global Are.na wall. saves returns only your personal moodboard; an optional key filters saves by source. Every response has entries, nextCursor, total, and coverage. Follow nextCursor with the same resource and key until it is null. total describes the catalog, not the number of search matches. Catalogs and personal saves can change during an export; pagination is not a frozen snapshot. Each asset entry contains the complete upstream record and an asset reference. Copy its resource, key, id, and page into readMoodsAsset. Keep the source and ID together: archive IDs can be negative, and an ID alone does not identify its catalog. The page is required to locate records in paginated archive sources. readMoodsAsset resolves available shot shards and preserves descriptions, attribution, source links, frames, transcripts, and shot/clip URLs. For Are.na blocks it returns the live block detail. Absent analysis stays absent; catalog availability does not mean every item has a transcript or original master. Literal text search:
This matches a case-insensitive phrase across stored catalog titles, curators, descriptions, details, tags/facets, work, and transcripts. It scans one catalog page at a time. Pass nextCursor as search.cursor to continue, even if the current page has no matches. Live Are.na block details and separate shot shards are not included in this text scan. To search all source catalogs, discover the sources and paginate each; omit scope to search the global Are.na catalog. Semantic search:
Are.na supports text, image, and both. Text mode means metadata embeddings, not literal full-text matching. For archive image embeddings use search: {type: "archives"}; optionally include a source slug as search.source. Archive search has no mode parameter. Semantic results are top-k candidates and have no pagination or completeness guarantee. Archive semantic results retain source, origin_id, r2_key, and job_key. These are provenance, not playable URLs: some identify packed objects. Use source catalog records for playable/downloadable media. Vector identities and catalog identities are separate; do not reinterpret a vector ID as a catalog ID. Not every indexed item is in a viewer catalog. Use searchArenaCorpus when a visual contact sheet is useful or to find neighbors with similarToBlockId. Preserve curator and artist attribution. Search tools require the existing Kino research entitlement and quota; browsing and reading follow the authenticated Moods viewer’s access rules. Normal MCP usage limits still apply.

​
Download media

Use the returned asset.originalUrl, videoUrl, previewUrl, or thumbnailUrl. Keep their meanings distinct; a preview or poster is not the original video. Original source links and archived clip links also remain in the record. Packed archive media receives a signed URL valid for 15 minutes. It supports GET, HEAD, and byte ranges without cookies. Re-read the asset to refresh an expired link. Treat signed links as temporary access grants and avoid publishing them. Raw api/archive/media/... values in record are internal locators; use the resolved asset URL for downloads. For a complete local export, preserve every returned record, source metadata, and asset reference, and follow every catalog cursor. Fetch detailed assets where you need shots or live block detail. This exports the data exposed by Moods; it does not export private infrastructure, embeddings, other users’ saves, or media not present in the catalog.

​
Prompt for your agent

Use Kino MCP through my signed-in Kino Desktop. Call getKinoInstructions, then browseMoods to discover sources. Search with searchMoods: literal for exact catalog text, semantic for visual or conceptual similarity. Continue every literal-search cursor, including empty pages. Read selected assets with readMoodsAsset, preserving resource/key/id/page, attribution, all metadata, and original/preview/thumbnail distinctions. Download through returned asset URLs; refresh expired packed-media links by rereading the asset. Use browseMoods with resource saves for my moodboard. Report catalog and search coverage honestly, and respect entitlement refusals.

​
Usage attribution

Successful MCP catalog calls record the authenticated account, tool, source/scope, record count, and pagination state in moods_catalog_access analytics events. Packed-media requests emit moods_media_request, attributed to the account that issued the signed link. The issuer is covered by the signature; changing it invalidates the grant. A shared link does not identify its eventual downloader. These events do not include search text, record contents, or signed URLs. Telemetry is best-effort, not a durable audit log. Media events describe requests (including HEAD and ranges), not completed downloads or confirmed bytes delivered. Direct external media URLs bypass this proxy and cannot be tracked here. Browse/read still require sign-in; research search additionally requires the existing Pro research entitlement and quota (with the existing staff exemption).