About Crazyrouter Mcp Server
Crazyrouter Mcp Server connects MCP clients to Crazyrouter for text chat, synchronous image generation, and live model discovery. It is for clients using Crazyrouter APIs through three focused tools.
Key features
- 01Text chat with supported models
- 02Synchronous image generation
- 03Live model discovery with filters
Use cases
- 01Send text prompts to chat models
- 02Generate a single image from a prompt
- 03Check models available to an API key
MCP support
Exposes chat, generate_image, and list_models as tools to MCP clients, using a Crazyrouter API key.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
4Saved tool list · updated 3w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| model | string | No | The AI model to use (default: gpt-5-mini). Examples: gpt-5, claude-opus-4-6, gemini-3-pro, deepseek-r1, llama-4-scout, qwen3-235b, grok-4 |
| messages | array | Yes | Array of chat messages with role and content |
| temperature | number | No | Sampling temperature (0-2). Lower = more deterministic, higher = more creative |
| max_tokens | number | No | Maximum number of tokens to generate |
Details
From the project READMECrazyrouter MCP: chat, image generation and model discovery
Practical guides / 实操入口
Updated 2026-10-12. This MCP server connects supported clients to Crazyrouter using your API key. The repository implements three tools; the gateway's complete catalogue is broader than these tools.
Tools
| Tool | API | Scope |
|---|---|---|
chat | POST /v1/chat/completions | Text messages for models listed for Chat Completions; default gpt-4o-mini |
list_models | GET /v1/models plus public endpoint metadata | Live models available to your key; optional category filter |
generate_image | POST /v1/images/generations | One synchronous image; default gpt-image-2; model-specific sizes |
Model discovery does not imply that this MCP implements each model's protocol. Video, music, transcription, TTS and Midjourney tasks require their own APIs. The old generate_video implementation sent video models to a chat endpoint and has been removed. Use the API documentation for those task APIs.
The server no longer silently replaces failed discovery with a hardcoded model list. Authentication and discovery errors are returned to the client. Categories use public endpoint metadata and name-based audio detection; unknown types appear under other.
Install from npm
Requires Node.js 22 or later.
Configure your MCP client with npx -y [email protected] and set CRAZYROUTER_API_KEY as shown below. The client starts the server automatically.
To build from source instead:
git clone https://github.com/xujfcn/crazyrouter-mcp.git
cd crazyrouter-mcp
npm ci
npm run build
Claude Desktop or Cursor
Create a key in the Crazyrouter console. Add the following to the client's MCP configuration, replacing the key locally:
{
"mcpServers": {
"crazyrouter": {
"command": "npx",
"args": ["-y", "[email protected]"],
"env": { "CRAZYROUTER_API_KEY": "YOUR_CRAZYROUTER_API_KEY" }
}
}
}
Claude Desktop uses its claude_desktop_config.json; Cursor supports .cursor/mcp.json. Restart the client after editing. Do not commit a configuration containing a real key.
VS Code
Use .vscode/mcp.json and the VS Code MCP schema:
{
"servers": {
"crazyrouter": {
"type": "stdio",
"command": "npx",
"args": ["-y", "[email protected]"],
"env": { "CRAZYROUTER_API_KEY": "${input:crazyrouter-api-key}" }
}
},
"inputs": [
{ "type": "promptString", "id": "crazyrouter-api-key", "description": "Crazyrouter API key", "password": true }
]
}
Example tool arguments
{"model":"gpt-4o-mini","messages":[{"role":"user","content":"Reply OK"}],"max_tokens":16}
Use only one of max_tokens and max_completion_tokens, according to the model's documented schema. Do not send optional temperature values to models that do not support them.
{"category":"image"}
{"model":"gpt-image-2","prompt":"A simple blue circle on a white background","n":1}
Image output may be a URL or base64 image. The tool returns both formats through MCP and does not report success for an empty result. Size constraints vary by model. Image requests incur the model's normal charges.
Prices and troubleshooting
- Current models and prices are the source of truth; there is no fixed model count or universal discount.
- HTTP 400: inspect the exact model ID, endpoint, size and other fields. Start with the minimal example.
- HTTP 401/403: check your key and permissions. HTTP 429: inspect quota and rate limits.
- Models requiring a native task API are rejected by the chat/image tools before inference.
- The fixed API base is
https://crazyrouter.com/v1; marketing parameters are never added to API requests.
Development
npm ci
npm run build
npm test
Tests use mocked HTTP responses through an MCP client/server connection. On 2026-10-12, live checks through the updated MCP passed for model discovery, a minimal gpt-4o-mini chat request and one gpt-image-2 image request. This does not establish that every model or media operation has been tested. Maintainers can run node tests/live-smoke.mjs; add --image to include one billable image request.
中文说明
本仓库支持文本聊天、同步图片生成、实时模型发现三个工具。模型目录会返回视频、音乐等类别,但不代表这些能力已在 MCP 中实现。旧版视频工具错误地调用聊天接口,已移除;请按 API 文档使用原生任务接口。
使用 Node.js 22+,按上面的配置通过 npx -y [email protected] 启动。也可以克隆仓库后运行 npm ci 和 npm run build,再将本地 dist/index.js 配置到 MCP 客户端。密钥通过 CRAZYROUTER_API_KEY 传入。不要将模型目录中的显示名称当作精确 API ID,也不要将推广参数添加到接口地址。
License
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.