SinoGate MCP Server
OfficialClick on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@SinoGate MCP Server compare DeepSeek and Qwen on the pros and cons of Rust vs Go"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
SinoGate MCP Server
Use DeepSeek, Kimi (Moonshot), Qwen, GLM, MiniMax, and other models in SinoGate from Claude Desktop, Claude Code, Cursor, Cline, and MCP-compatible clients. Requests use the OpenAI-compatible SinoGate API.
Each chat and comparison call uses your own SINOGATE_API_KEY. SinoGate bills those requests to your account.
Tools
sinogate_chat— Send a prompt or conversation to one model. Returns the completion and usage information.sinogate_list_models— List enabled models, capabilities, and current public USD prices. Filter by provider.sinogate_compare— Run one prompt against two to four models in parallel and compare answers, latency, and token usage.
Prices come from SinoGate's public model catalog. The authoritative pricing page is sinogate.ai/docs/models-pricing.
Related MCP server: universal-ai-agent-bridge
Requirements
Node.js 20 or later
A SinoGate account and API key from sinogate.ai
Claude Desktop
Add this server to claude_desktop_config.json:
{
"mcpServers": {
"sinogate": {
"command": "npx",
"args": ["-y", "@sinogate/mcp"],
"env": {
"SINOGATE_API_KEY": "sk-sg-your-key"
}
}
}
}Restart Claude Desktop after saving the configuration.
Cursor
Add the same server entry to your Cursor MCP configuration (~/.cursor/mcp.json).
Claude Code
claude mcp add sinogate -e SINOGATE_API_KEY=sk-sg-your-key -- npx -y @sinogate/mcpOther stdio MCP clients can launch npx -y @sinogate/mcp and provide the environment variable in their MCP server settings.
Configuration
Variable | Required | Default | Description |
| Yes | — | Your SinoGate API key. Chat and comparison requests are billed to your account. |
| No |
| API base URL. |
The server sends Authorization: Bearer <key> only to the configured SinoGate API host. It does not log the key. The list models tool reads public model and pricing data.
Run from source
npm install
npm run build
SINOGATE_API_KEY=sk-sg-your-key node dist/index.jsOn PowerShell, set the key for the current session with $env:SINOGATE_API_KEY = "sk-sg-your-key" before running Node.
License
MIT
Available Tools
3 toolssinogate_chatChat with a SinoGate modelB
Send a prompt or conversation to a SinoGate model such as DeepSeek, Kimi, Qwen, GLM, or MiniMax. Uses the configured SinoGate API key and returns the answer and token usage.
Inputs: model (required), prompt, messages, thinking, +2 more.
sinogate_compareCompare SinoGate modelsB
Run the same prompt against 2 to 4 SinoGate model IDs in parallel. Returns each answer, latency, and token usage; calls are billed to the configured account.
Inputs: models (required), prompt (required), thinking, max_tokens, +1 more.
sinogate_list_modelsList SinoGate models and pricesA
List currently available SinoGate models, modalities, thinking support, and public USD prices. Optionally filter by provider such as deepseek, kimi, qwen, glm, or minimax. Prices link to the official pricing page.
Inputs: provider.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool targets a clearly different action: single-model inference (chat), metadata discovery (list_models), and multi-model parallel inference (compare). The compare tool is explicitly bounded to 2-4 model IDs, so it does not overlap with chat in practice.
All three tools share a consistent `sinogate_` prefix, snake_case, and verb-led naming. Minor deviation: `chat` and `compare` are bare verbs without an object noun, unlike `list_models`, but the pattern is still predictable.
Three tools is lean but reasonable for a thin model-gateway wrapper: discover models, run one, run several. It is on the sparse side given the domain also implies modalities, streaming, and account usage.
The core inference and model-discovery lifecycle is covered, but obvious gaps remain: no streaming/chat-session handling, no embeddings or non-text modality invocation despite listing modalities, and no account balance or usage/billing lookup beyond per-call token counts.
Maintenance
Related MCP Connectors
List, compare, and call 68 LLMs with live pricing through one Chinese API gateway.
Connect MCP clients to 2,000+ AI models without managing provider API keys.
China MCP gateway hosting HTTP APIs for Doubao/Tongyi/Wenxin/Claude; weather, currency, holiday.
Chat, image, video, and audio models from many vendors behind one remote MCP server and one key.
Related MCP Servers
- AlicenseAqualityCmaintenanceEnables interaction with OpenAI-compatible model proxy APIs through MCP, supporting text generation, vision understanding, and image generation with automatic model discovery and routing.3MIT
- AlicenseAqualityBmaintenanceEnables any AI agent to connect to any OpenAI-compatible LLM provider through a unified MCP interface, with automatic failover, caching, cost tracking, and multi-model consensus.13MIT
- AlicenseAqualityBmaintenanceExposes 120+ Chinese frontier AI models (DeepSeek, Qwen, GLM, Kimi, Seedance, MiniMax, Hunyuan, CogVideoX) as MCP tools via one OpenAI-compatible endpoint: chat, image generation, free-model list, and a live first-party price leaderboard. Zero platform fee.7MIT
- AlicenseNot gradedqualityAmaintenanceEnables MCP-compatible applications to interact with local and cloud-hosted LLM models for model management, text generation, chat, embeddings, and web search/fetch.817 npmAGPL 3.0