AetherWave MCP is the Model Context Protocol (MCP) server for AetherWave Studio, aimed at MCP-compatible clients such as Claude Code, Cursor, Continue, Claude Desktop, and custom agents. Its core purpose is to let large language models call multiple flagship creative AI services through a single API key and a single credit pool, covering music, image, video, and audio processing.
In terms of main capabilities, thirty-three tools are provided by default. Music generation supports Suno V3.5 / V4 / V4.5 / V5 / V5.5. Image generation supports Grok Imagine, GPT Image 2, Seedream V4, Wan 2.7, Nano Banana 2, Flux 2, Qwen Image, and Z-Image Turbo. Image editing supports Grok Imagine I2I, Seedream V4 Edit, Flux Kontext, Wan 2.5 Spicy, Qwen Edit, and GPT Image 1.5. Image utilities include 2x / 4x upscaling, Recraft background removal (with fal BiRefNet v2 fallback), and Ideogram V3 Reframe. Video generation supports Grok Imagine (KIE+fal fallback), Wan 2.7, Hailuo 02, Seedance Pro/Lite, Kling 2.6 (audio), VEO 3.1, and Happy Horse. Video utilities include Atlas upscaler (1080p/2K), rembg u2netp background removal, and Luma Ray 2 Flash reframing. Audio mastering provides twelve genre/style presets, handled by the AetherWave Python service. There are also gallery tools (upload images, audio, and video from local files, URLs, or data URLs, and list saved works with pagination), a user manual PDF link, comic book tools (nine tools driving the Graphic Novel Engine, from characters, character references, scripts, storyboards, and lettered pages to PDF / EPUB / CBZ download links), and merchandise tools (four tools: apparel catalog,design-to-mockup, and after setting PRINTFUL_API_TOKEN, two more tools for creating and listing Printful store products, plus a tool for managing store listings).
In terms of usage, each generation tool submits a job and waits for a soft deadline of about 45 seconds. Fast jobs return the final URL in a single round trip; longer jobs return state: "running" and a taskId, and aetherwave_get_job can be used to poll all pipelines (image, video, music, mastering, comics, merchandise). Passing async: true to any tool returns immediately. Results are automatically saved to the Cloudflare R2 gallery, and URLs do not expire. For installation and configuration, Claude Desktop requires editing ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows) and then restarting; Cursor is added via Settings -> MCP -> Add new server; Continue (VS Code / JetBrains) is configured in ~/.continue/config.yaml; custom or programmatic clients can use the standard stdio MCP server, and any client that uses JSON-RPC 2.0 over stdio according to the MCP specification can connect.
Use cases include: having an agent generate a complete comic book from a single sentence, creating multi-shot videos with saved UGC characters while keeping faces and voices consistent, generating and listing Printful merchandise through conversation, and batch generating, editing, upscaling, reframing, and mastering various media content.
Limitations and notes: MCP clients usually give up on a tool call after about 60 seconds, so long jobs need polling; video rendering usually takes 1 to 8 minutes; comic book drawing takes about an hour, and slow steps return immediately and are checked with aetherwave_comic_status; script rewriting deletes all storyboards unless replaceExisting: true; storyboards rejected by content filtering should be rewritten rather than retried; only permanent images can be used as character references; imageUrl and referenceImages cannot be stacked; voice text must be byte-for-byte identical to maintain character consistency; generated speech may mispronounce words and still report success, so when unattended, it is recommended to transcribe and verify first; Printful charges per order, so you must confirm on your own that the retail price covers costs and that you have the rights to the assets. Printify supports labeling as coming soon.
