Lyria MCP: generate music with Lyria 3.5 from Claude

To use Lyria over MCP, connect a remote MCP server that runs Google's Lyria music model. On Sume's hosted server, the paid music_create tool turns a text prompt into a music track, and today every request resolves through Sume's Music Router, which picks Lyria 3.5.

The facts come from Sume's MCP overview, Music 1.0 and Music Router docs, the API reference and the hosted tool's code, read on 2026-09-28. The basics page says hosted MCP still works but is not the primary path today; from a backend, Music generation API with Lyria 3.5 covers the REST call.

Which Lyria model does music_create run?

music_create submits a Music 1.0 job (sume/music-1.0). Music 1.0 is retiring gradually: its routes keep working, but every request now resolves through the Music Router (sume/music-auto), which picks Lyria 3.5 today. The job's request.routed_model names the engine that ran, for example lyria-3.5. The docs describe Lyria 3.5 on Sume as full-length structured songs up to a few minutes, steered by the prompt.

The tool takes no model id, so you cannot pin lyria-3-pro over MCP. Here is how it compares with the REST route:

What does a music_create call take?

Every field goes inside payload, next to a required idempotency_key. prompt is required (1–5000 characters) and image_url is an optional public HTTPS image; a non-empty negative_prompt returns a 400, so put exclusions in the prompt. Music generation API with Lyria 3.5 lists the rest of the body.

{
  "idempotency_key": "bgm-product-teaser-001",
  "dry_run": true,
  "payload": {
    "prompt": "Warm lo-fi hip hop, 84 BPM, C minor. Dusty Rhodes, brushed drums. A 30-second track. Instrumental, no vocals."
  }
}

How do I run Lyria from Claude Code?

The steps are the same in Cursor or Codex once the server is added. For writing the brief itself, see AI music prompt examples.

• Add the server: claude mcp add --transport http sume https://mcp.sume.com/mcp, then claude mcp login sume.
• Turn Write on at consent. A read-only session (mcp:read) gets insufficient_scope from paid tools; an API-key session sees the full tool set.
• Call music_create with dry_run: true to preview admission and cost, then again without it to submit.
• Wait with jobs_wait, which holds at most 55 seconds per call, then read jobs_result. The audio is in result.artifacts[] where type is audio, typically audio/mpeg on media.sume.com.

How much does Lyria cost over MCP?

A music generation is listed at $0.125 per audio on API pricing, and the Music Router docs say every engine charges the same fixed Music price. The price does not change with prompt length or an image input. max_spend_usd caps a call when you send it.

What are the limits?

To lay the track under a clip, Generate music for a video with AI covers fitting it to the cut. Over MCP, keep these in mind:

• There is no length or seed parameter: steer length in the prompt, such as "a 2-minute track" or section markers like [0:00-0:30] Intro: …, and check what comes back.
• Hosted MCP cannot read files from your laptop, so an image_url must already be public.
• music_create makes music only, not speech; for a voiceover, see Claude text to speech over MCP.