Cue for AI

Cue speaks MCP.

Connect your AI to Cue and it works your catalog like you would — researching artists, keeping EPKs current, and building record pages with the music on them. Here's the whole flow.

The gist

Cue is an MCP server. Point an MCP client — Claude, or anything that speaks the protocol — at it, approve one prompt, and your AI can read and write your real workspace: the same EPKs and records you see in the browser. No export, no copy-paste, no second copy of the truth drifting out of date.

you driveYour AIClaude, or any MCP client
OAuth · one approval
the connectorCuecue.re/api/mcp
reads + writes
your catalogEPKs · Recordspress pages · audio · analytics

No copy-paste, no export. Your AI works on the same live workspace you see in the browser.

Connect it once

In your MCP client's connector settings, add a custom connector pointed at:

connector urlhttps://cue.re/api/mcp

The client discovers the rest on its own and shows you a Connect screen naming the workspace it's about to touch. Approve it and you're live — the grant is scoped to that one workspace, and you can pull it any time from your account.

The approval is real OAuth, not a pasted key. Cue never asks you to hand a secret to your AI, and the connection only ever reaches the workspace you approved.

The one idea: EPK, then records

Everything downstream rests on one distinction. An EPK is the artist's durable record — one per artist, the material your AI researches and keeps current: the artist name, the assets and photos, the credits and quotes. A record is one release presented, and it writes its own story — its hook, its press release, its note are the record's own words, not borrowed prose. The EPK lends the artist name and assets, and on the public page it backfills any kit field the record doesn't set itself.

The EPKthe base layer · one per artist

Your AI researches it and keeps it current. The durable material lives here once — the artist name, the assets, the record.

artist namecreditsattributed quotessync metadataassets + photoslinks
lends the name + assets · backfills what a record leaves blank
recordFor a writer
owns its story+ its own hook + press+ personal note+ one CTA: premiere
recordFor a sync supervisor
owns its story+ brief-matched cut+ one-stop flag+ CTA: license
recordFor radio
owns its story+ add date+ clean edit+ CTA: add

Each record writes its own story. The EPK lends the artist name and assets — and any kit field a record doesn't set still renders the EPK's value on the public page, so updating the EPK moves every un-overridden field at once.

The payoff: durable facts live once on the EPK — update it when a placement lands and every un-overridden field moves across your records — while each record page stays about the release: the story, every track playing, the record as MP3 or WAV, the press kit, the single ask.

What a full run looks like

A typical session, start to finish. You talk; your AI calls the tools.

  1. 1
    Researchyour AI digs into the artist, press, DSPs
  2. 2
    Build the EPKupsert_artist_epk — the durable one-sheet
  3. 3
    Build the recordcreate_pitch — the page, audio inline
  4. 4
    Send it yourselfthe link goes in your own mail
  5. 5
    Watch it landget_pitch_analytics — opens + plays
say to your AI
“Research Perfect Traveler — press, DSPs, credits — and build out their EPK. Keep it specific and un-hyped; don't invent any quotes.”
→ list_releases, then upsert_artist_epk (merges into the existing kit)
say to your AI
“Build the record page for From the Narrows — lead with the title track, write the press release in my voice from the kit, credits included — and give me the link and a short email to go with it.”
→ create_pitch, inheriting the EPK; the link comes back
say to your AI
“Which records got opened this week, and what did people actually play?”
→ get_pitch_analytics

Every tool, grouped

Grouped by job. You never call these by hand — your AI picks them — but here's a map of the main tools so you know what it can reach.

Research & EPK
  • list_releasesfind an artist's EPK id
  • get_press_kitread the current kit
  • upsert_artist_epkwrite research to the EPK — merges by default
  • resolve_listen_linksone streaming link → verified where-to-listen on every DSP
  • mine_bandcampcredits, notes and dates from a public page
  • suggest_similar_artistsan honest fans-of line
Records
  • create_pitcha record page with the audio inline
  • update_pitchenrich or edit one
  • add_photosphotos from links — the gallery + press package
  • list_pitchessee every record
  • get_pitchread one back
  • get_pitch_analyticswhat got opened, what got played
  • delete_pitchretire one
Catalog
  • update_track_syncbpm, key, isrc, one-stop and the rights a supervisor asks for

Sending it

Cue doesn't send mail. When the page is ready your AI hands you the link; you paste it into your own email, from your own address, with your own publicist's playbook — or let your AI draft the note and send it through whatever mail it already has access to. Every door is the page's own: a writer plays every track, downloads the record as MP3 or WAV, and takes the press kit at the foot without anyone granting anything.

The connector reaches your workspace, never your inbox. Cue holds no contact list and no sending identity, so there is nothing here to warm up, verify, or get blocked.

Good habits

  • Durable facts on the EPK, the story on the record. Assets, credits, and quotes live once on the EPK; each record writes its own hook and press release.
  • Specific over hyped. Real quotes, real numbers, real placements — never adjectives, never an invented attributed quote.
  • One page per record, usually. A per-recipient page gets its own link and its own analytics; make one when you have a short list and a real reason.
  • Read the page back before you send. Every name, date, and credit on it should trace to the kit.
  • Let analytics steer the follow-up. get_pitch_analytics shows opens and plays; a second note to someone who replayed the hook beats a blast.
Cue speaks MCP. · Cue docs