WaveSpeed MCP Server
WaveSpeed MCP server: an agent runs image, video, audio and 3D models, reads each model's input schema and quotes the price first
- 37 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Runs spend credits on the WaveSpeed account, and the description names no server-side cap
- The delete_predictions tool wipes up to 100 runs from history
Install
- Works in
Manual install
claude mcp add --transport http wavespeed https://api.wavespeed.ai/mcpHosted server; access is approved on the WaveSpeed page.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Runs spend credits on the WaveSpeed account, and the description names no server-side cap
- The delete_predictions tool wipes up to 100 runs from history
This is third-party code. Review the repository files before installing.
What it does
The server connects an agent to the WaveSpeed platform. It takes the model list from the live API catalog with a one-hour cache, so new models work the day they ship. The general run_model tool runs any model, and get_model_schema first shows which fields that model accepts. Invalid keys are rejected before submission. Five shortcut tools cover common jobs: generate_image, edit_image, generate_video, edit_video and extend_video. They use the same models as the WaveSpeed chat page, pick the endpoint from the inputs and fit the request to the model schema. What was changed is reported in the notes field. The estimate_only flag returns the price and the exact request without running it. In an input value, a string like "@./photo.jpg" uploads a local file and substitutes the hosted URL. Long jobs return an id, and the agent polls get_prediction until done is true: waiting is allowed up to 1800 seconds, and the task keeps running server-side if the connection drops. Separate tools show the balance, run history, charges and spend.
Who it is for. Designers and product teams who generate images and clips straight from Claude Code or Cursor and want to see the price before a run.
Good fit when
- You need a 16:9 hero image and want the result saved into the project as a link
- A photo's background from the project folder is swapped, with the file uploaded through an @path marker
- A still drawing becomes a short video, after a look at the cost estimate
- A finished clip is extended with new footage through extend_video
- At month end you compare spend by model and the list of runs for the period
Not a fit when
- Generation must run locally and for free: the server calls the paid WaveSpeed service
- Tasks about text, code or data analysis: the server only covers image, video, audio and 3D models
- Nobody can issue a billing key: list_billings and get_usage need a key with billing access
- The agent runs in ChatGPT or Claude on the web and must read files from your computer: the hosted variant cannot see them
Example request
Generate a 16:9 cyberpunk city skyline at golden hour, show me the price first and then run it.MCP
- Transport
- stdio, http
- Authentication
- key is optional
Environment variables
WAVESPEED_API_KEYsecret- WaveSpeed API key for local runs, taken from the keys page.
Limitations
Every run spends credits on the WaveSpeed account; the service decides the final amount, and the get_price estimate stays approximate. Local mode needs a WAVESPEED_API_KEY from the account page. The hosted server cannot read files on your computer, so pass public URLs or a temporary upload address from create_upload. The delete_predictions tool removes up to 100 runs from history with no way back. Whether WaveSpeed is reachable from Russia was not checked for this card.
How to disable. In Claude Code, remove the server with claude mcp remove wavespeed. For a connection through Claude on the web or ChatGPT, delete the key named MCP on the WaveSpeed API Keys page, and access closes at once.
FAQ
How do I connect WaveSpeed MCP to Claude Code?
Run claude mcp add --transport http wavespeed https://api.wavespeed.ai/mcp and approve access on the WaveSpeed page. The local variant installs with claude mcp add wavespeed -- npx -y @wavespeed/mcp@latest, and the key comes from WAVESPEED_API_KEY.
How much does a run cost?
get_price shows the cost for the same inputs without charging. The service decides the charged amount, which can differ from the estimate.
How is generate_video different from run_model?
generate_video picks the mode (text, frame, references) from the inputs and uses the chat page models, Seedance 2.5 by default. run_model runs any catalog model and needs exact fields from the schema.
What if a generation takes long?
The task returns an id, and the agent calls get_prediction with wait_seconds up to 1800. If the connection drops, the task keeps running on the server.
How do I revoke access granted through Claude on the web?
Open the API Keys page in WaveSpeed and delete the key named MCP with the app name. The connection ends at once.
Discussion
No questions or reviews yet.
Related
MCP serversRunAPI MCP server: a catalog of 160 models for images, video, music and speech, pricing, task runs and balance
Hosted Magic Hour MCP server: the agent generates and edits video, images and audio through the service API
An MCP server that edits images through Photopea: layers, text, effects and export, 46 tools for Claude Code and Cursor.
A local MCP server for editing and analyzing images: the agent crops frames, finds objects and reads text from photos
MCP server for image generation and editing: Gemini, OpenAI and Seedream with automatic prompt enhancement
An Excalidraw MCP server: the agent draws a diagram and adds keyframe animation while you watch it assemble live in the browser
README in short
The documentation describes two ways to run it: a hosted address on api.wavespeed.ai and a local package through npx. Both expose the same tools and differ in how they handle files. The server follows the design of the open @wavespeed/cli, and one login covers both tools.