OffStereo MCP Server

Use action=article with title, url and pasted text to draft a cited adaptation for review; this does not fetch the URL. Inspect its evidence and narration before action=review. Use action=intake with title and notes to organize a pasted document into an editable draft. Review all extracted identities and unassigned lines before action=review. Turn a supplied ordered set list and exact liner notes into a playable story without rewriting them. First use action=review with material, present the returned recordings, narration and issues for approval, then action=prepare with that reviewId after approval. To edit an owned story, action=revise takes sessionId and narration changes, returns a new review, and leaves the original unchanged. Reviews expire after 30 minutes. After preparation, export_video with sessionId and narration beatId creates a short narrated video; video_status with jobId reports progress and signed-in browser download links. Downloads require the same owner to sign in; these links are not public or bearer download URLs. Use credits with sessionId for the saved transcript, source credits, recording identities and owner-only return path. Does not publish or start playback. Music rights are not granted and narration export does not include music.

Discover genre-aware new releases, strong public-archive matches, or popular OffStereo shows. Call exactly once per listener request: put every positive genre into the same genres array, use [] when none were requested, and avoid one call or card per genre. query carries the listener's full wording, exclusions, counts, and no-padding constraint. Results remain relevant rather than padding a requested genre or archive search with unrelated rows. Dated music-history requests use music_today and are not a discover mode. playableOnly limits results to music that plays in ChatGPT. allowBuild and allowSave preserve a read-only request. The result card already includes song playback; a separate one-song card is for an explicit request for a dedicated song card.

Explore documented sample lineage. Pass artist + track for one song, or topic for one multi-song narrated lineage story. OffStereo only returns connections it can source and withholds uncertain matches rather than guessing.

Create one playable OffStereo card for a direct standalone-song or narrated-story request. Discovery and Music Today already play their songs, so keep those results in their original card unless the listener asks for dedicated controls. experience=single_track returns exactly one recording with no narration or library save; pass artist plus track. experience=story progressively creates a sourced story; mode=fresh starts a new build. Pass artist plus album for an exact record and trackCount for an exact requested length. Spotify, Apple Music, and SoundCloud URLs are story seeds unless the listener requests only the recording. Playlist requests use create_playlist. existingSessionId opens a selected library story; get_session is reserved for card polling. The mounted card owns live progress and playback. Assistant prose should avoid restating transient stages such as building, recording, or still working because they become stale as the card updates, and should wait for Ready before stating final counts. A tool call opens the card without starting playback.

Manage the signed-in listener's own OffStereo stories or service playlists. Removing an OffStereo story is a recoverable archive operation, and restore brings it back. Removing an owned Spotify playlist means remove from library because Spotify does not permanently delete playlists; the operation follows an explicit confirmation turn. Apple Music's public API does not expose playlist deletion, so that case returns an honest unsupported handoff. confirm=true records explicit confirmation of the named item and operation.

Call exactly once for a 'today in music' or dated music-history request. One response already combines album anniversaries, artist birthdays and memorials, current releases, current US song-chart context, and related playable OffStereo stories; use that one combined call rather than separate history, releases, charts, or shows calls. dateMode=today uses the listener's current day and omits date; dateMode=explicit pairs with a named YYYY-MM-DD date. limit is the maximum total across the combined response, playableOnly limits rows to music that plays in ChatGPT, and allowBuild=false plus allowSave=false preserves a read-only request. This is the only dated-music tool in the current schema.

Call once to return the signed-in listener's OffStereo library overview, currently including their completed, unarchived shows. This is the primary personal-library tool.

Create one playlist from a natural-language idea or explicit track list. Call exactly once per listener turn: do not inspect the result and make a corrective second call or stack another playlist card in the same response. previewOnly=true returns one authoritative playlist that can be heard before saving and needs no music-service connection. After explicit approval, previewOnly=false with the approved tracks and the same concise editorial name creates one private playlist in the listener's active music service. Pass service only when the listener explicitly names Spotify or Apple Music. requiredArtists carries every artist the listener named as required. allowSecondPlaylist is true only for an explicit request for a separate additional playlist. Sign-in recovery preserves the exact request and pendingActionId for a safe retry.

Prepare one finished OffStereo story to become a private Spotify episode after the listener explicitly asks. OffStereo packages its completed intro, ordered narration, sign-off, branded title, show notes, OffStereo cover, chapters, and song timeline, then returns an honest handoff: the episode is not in Spotify until the listener finishes with Spotify's Save to Spotify helper on their own computer. Use the active or selected story's sessionId. This is for a narrated story, not a playlist or an unfinished build.

Explicitly save the matched music from one listening session as a private playlist in the listener's active music service. Pass service only when the listener explicitly names Spotify or Apple Music. Unmatched songs are excluded and reported. If sign-in is needed, OffStereo preserves the request for one safe retry.

Application polling for an already-mounted OffStereo card. It refreshes one PlayableSession in place. A selected library story opens through make_playable(existingSessionId), while playback and contextual follow-ups use steer_session.

Control and contextual follow-ups for the active player: pause, resume, skip, play the next song, append one exact next song, adjust the music level, explain why a song is present, open an explicitly requested documented sample lineage, or return to the story. The result updates the mounted card in place. For an AI-DJ transition to a newly chosen record, pass queueTrack with exact artist and title; OffStereo resolves it and appends it to this same widget so the current song can finish and auto-advance. Keeping the transition in this queue avoids a second make_playable call. Use instruction='Queue this as the next song without interrupting the current song' to queue ahead, or 'Play this next now' to append and immediately advance. A why-this-song question preserves the queue. sessionId identifies the active session, and activeItemId plus positionMs preserve exact continuity when the card supplies them.

Get a finished OffStereo story as ONE downloadable MP3 — the narrated story, without the records. Use this on any surface where a player card cannot render: send it as a WhatsApp or Telegram voice note, attach it to an email, play it over a phone or speaker, or hand it to someone as a file. The URL is public and needs no OffStereo account, so it can be forwarded to anyone. It is narration only, which is why it is freely shareable — say so rather than promising the songs; the songs are at webUrl. Prefer this over reading a summary aloud in your own synthetic voice: this is the produced story, already voiced. Only finished shows have audio — poll get_show until status is ready.