# MCP server

> Use 500+ Raywake image, video and audio models from Claude, ChatGPT, Claude Code, Cursor and any MCP client, with the hosted connector or the open-source npx package.

Canonical: https://raywake.si/docs/mcp
Last updated: 2026-11-02

The Raywake MCP server lets an AI assistant search the catalog, read a model’s inputs, quote a generation and run it after you approve the price. Connect the hosted server with nothing to install, or run the open-source [@raywake/mcp](https://www.npmjs.com/package/@raywake/mcp) package locally.

## Hosted connector

Add this URL to your assistant and sign in with your Raywake account. You choose a spending limit per generation when you connect, and you can disconnect at any time in settings.

MCP server URL:

```
https://mcp.raywake.com/v1
```

- **Claude.ai and Claude Desktop:** Settings → Connectors → Add custom connector, paste the URL, then connect.
- **ChatGPT:** Settings → Connectors → Create (developer mode), paste the URL, then sign in.
- **Claude Code:** run the command below, then `/mcp` to sign in.

Claude Code:

```
claude mcp add --transport http raywake https://mcp.raywake.com/v1
```

Prefer an API key over browser sign-in? Pass it as a header instead:

Claude Code with an API key:

```
claude mcp add --transport http raywake https://mcp.raywake.com/v1 \
  --header "Authorization: Bearer $RAYWAKE_API_KEY"
```

## Local package

The npm package runs over stdio and needs Node.js 20 or later and an [API key](https://raywake.si/docs/authentication) with `models:read`, `generate`, `jobs:read` and `wallet:read`.

Claude Code:

```
claude mcp add raywake -e RAYWAKE_API_KEY=YOUR_RAYWAKE_API_KEY -- npx -y @raywake/mcp
```

Cursor (`~/.cursor/mcp.json`), Claude Desktop (`claude_desktop_config.json`) and Windsurf (`~/.codeium/windsurf/mcp_config.json`) use the same entry:

mcp.json:

```
{
  "mcpServers": {
    "raywake": {
      "command": "npx",
      "args": ["-y", "@raywake/mcp"],
      "env": { "RAYWAKE_API_KEY": "YOUR_RAYWAKE_API_KEY" }
    }
  }
}
```

## Tools

**MCP tools**

| Name | Type | Description |
| --- | --- | --- |
| `search_models` | free | Search the catalog by text, category or modality (image, video, audio). |
| `list_categories` | free | List model categories. |
| `get_model` | free | Read a model’s input schema and starting price in USD. |
| `quote` | free | Price an exact model and input. Returns quote_id, price_usd and expires_at. |
| `generate` | paid | Run the approved quote with the same model and input. The only tool that spends balance. |
| `get_job` | free | Read a generation’s status and outputs, optionally waiting up to 45 seconds. |
| `list_jobs` | free | List your recent generations. |
| `get_account` | free | Read your available and reserved balance in USD. |
| `get_playbook` | free | Read the short workflow and spending rules. |

## What costs money

Only `generate` spends your balance. The assistant quotes first, shows you the price in USD and asks before it generates. A generation can never cost more than its quote. Set your client to ask for confirmation before paid tools, too.

Images and videos you pass as inputs must be public HTTPS URLs. The local package cannot upload files from your computer yet.

## Try asking

- “Find video models for a product ad and compare their starting prices.”
- “Quote a 5-second Kling 4.0 video of a burger assembling itself. Ask me before generating.”
- “I approve that price. Generate it and show me the result.”
- “What is my Raywake balance?”

Source code, issues and changelog: [github.com/oguzhankayan/raywake-mcp](https://github.com/oguzhankayan/raywake-mcp). Listed in the official MCP Registry as `si.raywake/mcp`.

[llms.txt](https://raywake.si/docs/llms)

## Related

- [API documentation](https://raywake.si/docs)