Run your label from Claude.

The release-ops layer shouldn't only live behind a UI.

Most label operators already keep an AI client open next to their inbox. Claude, ChatGPT, Cursor. They write replies in it, summarize emails, sketch out plans for the week. Then they switch tabs to do the actual work, which usually means clicking around five tools to find the thing they were just thinking about.

Model Context Protocol changes the shape of that. An AI client that speaks MCP can call tools directly, see live data from your systems, and execute real actions on your behalf. Labelflow's MCP server makes the AI you already use a working surface on top of your label, not a side panel to copy notes into.

Eighteen tools, two surfaces.

The MCP server exposes thirteen read tools and five write tools today, with more landing as the beta runs. Read tools let the AI pull live state out of your workspace; write tools let it act, with the same permission rules as the operator app.

Catalog & releases

List releases, get a specific release with tracks and metadata, check status. Create and update releases, update tracks, kick off a catalog import. The catalog operations your roadmap actually runs on.

Artists & contacts

List artists, get an artist's full profile and history, list their tracks across the catalog. The roster surface mirrored from the operator app, readable in conversation; artist writes land as the beta runs.

Communications

List communications by artist or release, see what went out, what's drafting and what's waiting on approval. Drafting new comms from the AI client lands as the beta runs; sends always keep the operator approval gate.

Deliverables & analytics

List what's outstanding, what's blocking, what was missed - and mark deliverables done. Read analytics for any release, the whole catalogue, or the label. The dashboards that drive the weekly review, queryable in conversation.

The AI you already use becomes a label OS.

A few of the things you can ask today, written the way operators actually phrase them:

OAuth, the standard way in.

Click connect inside your AI client, sign in to Labelflow, scoped consent, done. Every connected app is visible in your settings and revocable in one click. Everything goes through the same permission model as the operator app: workspace-scoped, role-aware, fully auditable. Personal access tokens for scripts and header-based clients are planned.

All actions logged. Every read and write tool call leaves a trail in your workspace activity log, attributed to the operator and the client. You see what the AI did, exactly when, with what arguments.

What MCP won't do.

Same AI policy as the rest of Labelflow. The MCP server inherits every restriction, not just the convenient ones.

No predictive A&R scoring.

Even via MCP, no "this artist is about to break" forecasts. AI helps you read what exists; A&R calls stay with you.

No bulk sending without approval.

Write tools can draft and queue. Final send still goes through the operator's approval, by default. Auto-send is opt-in per template, just like in the operator app.

No third-party model training.

Your workspace data, your AI client, your decision. Labelflow doesn't sell aggregated MCP traffic to model labs. AI policy applies end-to-end.

No silent destructive actions.

Deletes, status transitions, irreversible moves require explicit confirmation by default. The AI can't quietly empty a release pipeline because the prompt was ambiguous.

Operators who already think in AI clients.

If you keep Claude open all day, work in Cursor, or run your label through ChatGPT prompts and copy-paste rituals, MCP is the version of Labelflow that fits how you already work. If you mostly use the operator app and prefer it that way, MCP changes nothing for you. It's a parallel surface, not a replacement.

Bundled with both tiers (Label, Studio). No add-on cost. See pricing →

Live in beta.

The MCP server is live for beta workspaces now: connect Claude, ChatGPT, or Cursor from your settings and it works against your real catalog. The tool surface keeps growing through the beta, always inside the AI policy. See the full roadmap →