关于 SentioGate FinOps & LLM Router
BYOK AI 路由器和 FinOps 控制平面:审计 Token 浪费、检查工具安全性并排列模型
配置
远程服务无需本地安装。选择你使用的客户端,复制配置或使用一键安装。
https://api.sentiogate.dev/v1/mcpclaude mcp add --transport http sentiogate-sentiogate https://api.sentiogate.dev/v1/mcp{"mcpServers":{"sentiogate-sentiogate":{"type":"http","url":"https://api.sentiogate.dev/v1/mcp"}}}在 Cursor 中打开{"servers":{"sentiogate-sentiogate":{"type":"http","url":"https://api.sentiogate.dev/v1/mcp"}}}在 VS Code 中打开{"mcpServers":{"sentiogate-sentiogate":{"type":"http","url":"https://api.sentiogate.dev/v1/mcp"}}}{"mcpServers":{"sentiogate-sentiogate":{"serverUrl":"https://api.sentiogate.dev/v1/mcp"}}}命令在终端中运行;JSON 粘贴到对应客户端的 MCP 配置后重启。需要授权的服务会在首次连接时提示登录。
工具
README 原文English
来自项目 READMEREADME 暂无中文译文,以下为英文原文;中文介绍见上方。
@sentiogate/mcp
Official Model Context Protocol (MCP) server for SentioGate — Zero-markup BYOK LLM router & FinOps control plane.
Expose real-time LLM cost analytics, token-waste auditing, task-based model recommendations, and tool-safety inspection directly inside Claude Code, Cursor, Cline, OpenClaw, and Google Gemini / Search (Connected Apps).
Features (Exposed MCP Tools)
sentiogate_audit_waste— Scan past completion logs (JSON, JSONL, or CSV) to calculate dollar waste, prompt-cache misses, and potential savings with SentioGate.sentiogate_get_finops_ledger— Retrieve live spend, cache savings, and remaining 5-hour Community Pool quota.sentiogate_recommend_model— Classify prompt difficulty and receive optimal model + provider fallback chain based on empirical benchmarks and live TTFT.sentiogate_guard_check— Pre-flight check terminal commands and tool calls against SentioGate Armor safety rules (credential leak, destructive deletions, reverse shells).sentiogate_optimize_payload— Simulate prompt compaction and tool-output trimming to preview token savings before model invocation.
Quick Setup
1. Claude Code
Run in your project directory:
claude mcp add sentiogate -- npx -y @sentiogate/mcp
2. Cursor / Cline (mcpServers config)
Add to your mcpServers configuration (settings.json or ~/.cursor/mcp.json):
{
"mcpServers": {
"sentiogate": {
"command": "npx",
"args": ["-y", "@sentiogate/mcp"],
"env": {
"SENTIOGATE_API_KEY": "sg_live_your_api_key_here"
}
}
}
}
3. Remote HTTP Endpoint (Google Connected Apps / Gemini)
For remote MCP environments, point directly to the edge HTTP endpoint:
- Server URL:
https://api.sentiogate.dev/v1/mcp - Transport: Streamable HTTP / SSE (MCP
2024-11-05) - Headers:
Authorization: Bearer <your_key>(optional, public tools work unauthenticated)
License
Apache-2.0 © SentioGate Team
评论
一个开放协议,规定了 AI 应用如何以统一方式连接外部工具与数据源。客户端负责与模型对话,MCP 服务负责提供工具、资源和提示模板,两者通过标准消息通信,同一个服务可以被不同客户端复用。