WaveSpeedAI
← Blog/Workflows

Connect WaveSpeed to Claude, ChatGPT and Cursor with One URL

Add the hosted WaveSpeed MCP server to Claude, ChatGPT, Claude Code and other MCP clients with a single URL, then search models, check prices and generate images and video from the chat.

By WaveSpeedAI6 min read
#mcp#wavespeed-mcp#claude#chatgpt#cursor#claude-code#image-generation#video-generation

Most AI assistants can write a prompt for an image or a video. Fewer can run it, and fewer still can tell you what it will cost first. The WaveSpeed MCP server closes that gap: connect it once and Claude, ChatGPT or your editor can look through the live model catalog, check a price, generate the result and bring back the link, all inside the conversation you were already having.

The hosted server needs no install. It is one address:

https://api.wavespeed.ai/mcp

What can the assistant do once it is connected?

The server exposes a small set of tools, and the assistant chooses between them as the conversation needs:

  • Search the catalog. Find models by name, description or type, such as text-to-image or image-to-video. New models appear as soon as they are on wavespeed.ai, with no update on your side.
  • Read a model’s real inputs. Before running anything, the assistant looks up which inputs the model accepts, which are required and what the defaults are.
  • Quote a price. A price check that never charges, so you can see the cost before you approve a run.
  • Run a model. Image, video, audio and 3D models run from the same tool. When a generation takes longer than a chat turn, the assistant gets an ID back and checks on it again later.
  • Check your balance.
  • Bring in a file. Edits and animations need a source file; see the files section below.

How do I connect it?

The address is the same everywhere: https://api.wavespeed.ai/mcp. Clients that can open a browser sign you in and ask you to approve the connection. Clients that cannot, or that you configure by file, send an API key from the API Keys page as a bearer token instead.

Claude on the web

Add a custom connector, enter the address and approve the connection on the WaveSpeed page that opens. Claude lists the tools and asks before it uses them.

ChatGPT

Add a custom connector (this needs Developer Mode), enter the same address, choose OAuth and approve the connection.

Claude Code

claude mcp add --transport http wavespeed https://api.wavespeed.ai/mcp

The first time it is used, Claude Code opens your browser to approve the connection. To skip the browser and use an API key:

claude mcp add --transport http wavespeed https://api.wavespeed.ai/mcp \
  --header "Authorization: Bearer $WAVESPEED_API_KEY"

Cursor

Use the one-click link, or add this to your mcp.json:

{
  "mcpServers": {
    "wavespeed": {
      "url": "https://api.wavespeed.ai/mcp",
      "headers": { "Authorization": "Bearer YOUR_WAVESPEED_API_KEY" }
    }
  }
}

VS Code

From a terminal:

code --add-mcp '{"name":"wavespeed","type":"http","url":"https://api.wavespeed.ai/mcp"}'

Or put this in .vscode/mcp.json:

{
  "servers": {
    "wavespeed": {
      "type": "http",
      "url": "https://api.wavespeed.ai/mcp",
      "headers": { "Authorization": "Bearer YOUR_WAVESPEED_API_KEY" }
    }
  }
}

Any other client that supports remote MCP servers

Point it at the address and, if it has a place for request headers, send Authorization: Bearer YOUR_WAVESPEED_API_KEY. Create the key at wavespeed.ai/accesskey from any device, which also makes this the option for servers, containers and CI where no browser is available.

On your own machine instead

The same tools run locally as an npm package, for clients that start a command rather than connect to a URL (Claude Desktop, Cline and others). It signs in with WAVESPEED_API_KEY, or with the login the WaveSpeed CLI stores when you run wavespeed login.

{
  "mcpServers": {
    "wavespeed": {
      "command": "npx",
      "args": ["-y", "@wavespeed/mcp"],
      "env": { "WAVESPEED_API_KEY": "YOUR_WAVESPEED_API_KEY" }
    }
  }
}
claude mcp add wavespeed -- npx -y @wavespeed/mcp

For Codex, the WaveSpeed plugin bundles this local server: run codex plugin marketplace add WaveSpeedAI/codex-plugin-wavespeed-cli, then install WaveSpeed from the /plugins list. Per-client details are in the MCP guide, and the other ways to use WaveSpeed from agents and code are on the integrations page.

What does a session look like?

Ask in plain language and let the assistant pick the tools:

  • “What’s my WaveSpeed balance?”
  • “Find an image model that is good at product shots, tell me what a 2k image costs, then generate one: a ceramic mug on a wooden desk, soft morning light.”
  • “I have a still at this URL. Find an image-to-video model, quote a five-second clip, and run it with a slow push-in.”
  • “That video was still running earlier. Check whether it has finished and give me the link.”

Models differ in what they accept. Because the assistant reads each model’s inputs before it runs, and the server checks what it is about to send against them, a misspelled or invented parameter is caught up front instead of turning into an unexpected result.

Is it safe to give an assistant access to my account?

You stay in control at three points.

  • Approval. Most clients ask before a tool that spends credits runs. In Claude, the tool that runs models is listed as a write action and asks for approval by default.
  • A dedicated key. Each connection gets its own API key, named after the app. It appears on your API Keys page like any other key, and deleting it disconnects that app immediately.
  • Prices first. Ask for a quote before a run, especially for video, where cost depends on duration and resolution.

How do I give it a file?

Image edits and image-to-video need a source file. The simplest way is a public URL. Beyond that:

  • A tool that can make HTTP requests, such as a coding agent with a shell, can ask WaveSpeed for a temporary upload address, send the file there and use the link it gets back.
  • In ChatGPT, attach the file to the chat and the server imports it.
  • With the local server, a path such as @./photo.jpg uploads the file from your computer.

Where to go next

FAQ

What is the WaveSpeed MCP server?

A connector that lets an AI assistant use WaveSpeed directly: search the live model catalog, read a model's real inputs, quote a price, run image, video, audio and 3D models, and pick up long generations later. It is available as a hosted URL and as a local npx command.

Do I need to install anything?

No, not for the hosted server. Add https://api.wavespeed.ai/mcp as a connector in a client that supports remote MCP servers and approve the connection. The local server (npx -y @wavespeed/mcp) is there for clients that start a command on your machine.

How does sign-in work, and how do I disconnect?

Approving the connection on the WaveSpeed page creates a dedicated API key named after the app, for example MCP · Claude. The assistant uses that key to run models on your account. To disconnect, delete the key on the API Keys page.

Can the assistant spend my credits without asking?

Running a model uses your WaveSpeed credits, and most clients ask for approval before a tool that does so runs. You can also ask the assistant to quote the price first; the price tool never charges.

How do I give it a file to edit or animate?

Pass a public URL. In a tool that can make HTTP requests, the assistant can ask WaveSpeed for a temporary upload address and send the file there. In ChatGPT you can attach the file to the chat. The local server also accepts a path to a file on your computer.

Share