WaveSpeed MCP¶

Run any image, video, audio, or 3D model on the live WaveSpeed catalog from any MCP client. The official vendor server ships no hardcoded model list — the catalog comes from the live API, so new platform models work the day they ship. Every generation goes through the same run_model verb, with get_model_schema for real input schemas, get_price for quotes before spending, and get_prediction for async results.

Server type: Local (npm, stdio) with Dockerfile
Auth: WAVESPEED_API_KEY (or wavespeed login)
Package: npx -y @wavespeed/mcp (MIT)
Tools: 7
Registry: ai.wavespeed/mcp
Pricing: Pay-as-you-go credits (wavespeed.ai/accesskey)
Category: Media Generation
Built by: WaveSpeed AI (official)

Why This Matters for Operators¶

Marketing operators constantly need hero images, product renders, video clips, and audio assets, and every generation model has different inputs, costs, and quality tradeoffs. WaveSpeed MCP gives an agent one interface to the whole catalog: the agent can list models, read a model's real input schema before calling it, quote the price before spending, and check account balance — the same pre-spend discipline operators apply to ad budgets, applied to media spend.

The honesty mechanics matter for automation: get_price explicitly names the inputs the quote was blind to (unpriced_inputs) instead of presenting a formula floor as the real price, and local-file upload uses an explicit @path marker so inputs are never silently mutated.

Tools & Capabilities¶

Tool	What it does
`list_models`	Search the live catalog by text or modality
`get_model_schema`	A model's real input schema — required fields, properties, defaults
`run_model`	Submit and wait; `@path` inputs upload automatically; returns output URLs
`get_price`	Cost estimate with `unpriced_inputs` / `at_base_price` disclosure
`get_balance`	Account credit balance
`upload_file`	Local file to hosted URL (24h content-hash dedupe)
`get_prediction`	Recover status and outputs of any run by id

Long-running jobs keep running server-side past the wait limit — the error names the prediction id, and get_prediction picks it up.

Installation¶

claude mcp add wavespeed -- npx -y @wavespeed/mcp

Claude Desktop / other clients:

{
  "mcpServers": {
    "wavespeed": {
      "command": "npx",
      "args": ["-y", "@wavespeed/mcp"],
      "env": { "WAVESPEED_API_KEY": "wsk_..." }
    }
  }
}

Configuration¶

Auth resolves from WAVESPEED_API_KEY, or from the CLI's stored login (wavespeed login) — one login covers both tools. Keys are created at wavespeed.ai/accesskey. Repository: github.com/WaveSpeedAI/mcp-server (MIT). Listed on Glama with Quality A.

Business Relevance¶

Marketing operators generate campaign assets — hero images, product shots, short video clips — directly in the agent session
Content teams produce image/video/audio variants across models without leaving the assistant
E-commerce operators create product visuals and social creatives at scale from one catalog
Automation builders fold media generation into workflows with pre-spend price quotes and balance checks

Integration with CorpusIQ¶

WaveSpeed is the creative output layer; CorpusIQ is the business-data layer. A marketing operator can pull campaign performance from CorpusIQ (GA4, ad platforms, CRM), decide which creative to produce next, and generate it through WaveSpeed — then attribute the resulting content's performance back through the same CorpusIQ connectors. The price-quote-before-spend pattern in WaveSpeed's tools mirrors the spend visibility CorpusIQ gives across ad and email platforms.

Limitations¶

Media generation costs accrue per run; agents need a credit balance to execute
get_price quotes are estimates — unpriced_inputs names what the quote could not see
Output URLs are hosted by WaveSpeed; long-term storage is not guaranteed
Requires an account and API key; no anonymous free tier documented