cpzai-mcp-server
MCP server for CPZAI platform, enabling AI agents to manage trading strategies, run backtests, route orders across brokers, and access portfolios, risk analytics, and market data through natural language.
README
<h1 align="center">CPZAI MCP Server</h1>
<p align="center"> <strong>Model Context Protocol server for AI agent access to <a href="https://ai.cpz-lab.com">CPZAI</a></strong> </p>
<p align="center"> <a href="https://github.com/CPZ-Lab/cpzai-mcp-server"><img src="https://img.shields.io/badge/language-TypeScript-blue.svg" alt="TypeScript"></a> <a href="https://github.com/CPZ-Lab/cpzai-mcp-server/actions/workflows/ci.yml"><img src="https://github.com/CPZ-Lab/cpzai-mcp-server/actions/workflows/ci.yml/badge.svg" alt="CI"></a> <a href="https://modelcontextprotocol.io/"><img src="https://img.shields.io/badge/MCP-v1.0-green.svg" alt="MCP v1.0"></a> <img src="https://img.shields.io/badge/transport-Streamable_HTTP-orange.svg" alt="Streamable HTTP"> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License: MIT"></a> </p>
A production-grade Model Context Protocol (MCP) server that exposes CPZAI capabilities — strategy management, backtesting, multi-broker order routing (Alpaca / Interactive Brokers / FIX), portfolios, risk analytics, market data — as tools for any MCP-compatible AI agent (Claude, Cursor, GPT, etc.).
The hosted endpoint at https://mcp.cpz-lab.com/mcp is the only supported way to connect. This repo exists for transparency: anyone wiring their broker API keys through an AI agent should be able to read the exact code that handles those keys.
Quick Start
You'll need a CPZ API key + secret. Create one at ai.cpz-lab.com/settings/api-keys.
Cursor
Add to .cursor/mcp.json:
{
"mcpServers": {
"cpzai": {
"url": "https://mcp.cpz-lab.com/mcp",
"headers": {
"X-CPZ-Key": "your_cpz_key",
"X-CPZ-Secret": "your_cpz_secret"
}
}
}
}
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"cpzai": {
"url": "https://mcp.cpz-lab.com/mcp",
"headers": {
"X-CPZ-Key": "your_cpz_key",
"X-CPZ-Secret": "your_cpz_secret"
}
}
}
}
Claude (web / mobile) via OAuth
The server speaks OAuth 2.1 + PKCE at https://mcp.cpz-lab.com/.well-known/oauth-authorization-server. Use Claude's "Connect a server" flow and point it at https://mcp.cpz-lab.com/mcp.
Architecture
Agent (Cursor / Claude / GPT / custom)
│
│ Streamable HTTP + X-CPZ-Key / X-CPZ-Secret
▼
mcp.cpz-lab.com
│
▼
Node.js MCP server (stateless) ← this repo
│
│ per-request API key validation
▼
CPZ Platform REST API
Design decisions:
- Stateless transport. Each
POST /mcpcreates a fresh server instance — no session tracking, scales horizontally without coordination. - Thin protocol adapter. Validates the API key, then proxies tool calls to the CPZ REST API. Zero business logic in the MCP layer; what you see in
src/is what runs in production. - Key auth. Agents authenticate with CPZ platform API keys (
X-CPZ-Key/X-CPZ-Secret) — the same keys used by the REST API and thecpzPython SDK. - Paper-only by default. Live trading on a strategy requires explicit per-strategy promotion in the platform.
Available tools
18 tools exposed today, organized by domain.
| Tool | Description | Read-only |
|---|---|---|
| Strategies | ||
list_strategies |
List trading strategies with filtering | ✓ |
get_strategy |
Get a specific strategy by ID | ✓ |
create_strategy |
Create a new trading strategy | |
update_strategy |
Update an existing strategy | |
| Backtests | ||
get_backtest_results |
List backtest run results | ✓ |
| Orders & trading | ||
list_orders |
List trading orders with filtering | ✓ |
place_order |
Place a new trading order | |
list_positions |
List current portfolio positions | ✓ |
sync_portfolio |
Trigger portfolio sync across brokers | |
list_accounts |
List connected trading accounts | ✓ |
| Market data | ||
get_market_data |
Real-time quotes (price, bid/ask, volume) | ✓ |
| Risk | ||
compute_risk |
Compute fresh risk snapshot (VaR, Sharpe, drawdown) | |
list_risk_snapshots |
List historical risk snapshots | ✓ |
| Execution | ||
execute_strategy |
Execute a strategy on the Python backend | |
| Webhooks | ||
list_webhooks |
List webhook subscriptions | ✓ |
create_webhook |
Subscribe to platform events | |
delete_webhook |
Remove a webhook subscription | |
| User | ||
get_profile |
Get authenticated user profile | ✓ |
Local development
npm install
export CPZ_API_BASE_URL=https://api.cpz-lab.com
export CPZ_SERVICE_KEY=your_service_key
npm run dev
# server starts on http://localhost:3001
Tests + type check
npm test # single run
npm run test:watch # watch mode
npx tsc --noEmit # type check
Build
npm run build # compiles TS → dist/
npm start # runs the compiled server
Docker
docker build -t cpzai-mcp-server .
docker run --rm -p 3001:3001 \
-e CPZ_API_BASE_URL=https://api.cpz-lab.com \
-e CPZ_SERVICE_KEY=your_service_key \
cpzai-mcp-server
Security model
- All connections HTTPS with TLS 1.3 at the edge.
- API credentials are forwarded in HTTP headers and never logged.
- Per-request API key validation against the platform's
api_keystable. - All data access is user-scoped via the
user_idderived from the validated key. - OAuth tokens (when used) are held in server memory only — never persisted.
See the privacy policy for the full data-handling story.
Self-hosting
We don't currently support self-hosting. Use the hosted endpoint at https://mcp.cpz-lab.com/mcp with your CPZ API key. The hosted server is the same code in this repo plus AWS infrastructure (ALB, ECS Fargate, WAF, secrets) that's not part of this repo.
If you have a strong reason to self-host (e.g. air-gapped trading desk), open an issue and we'll talk.
Contributing
Issues and PRs welcome. Before submitting:
npx tsc --noEmitmust passnpm testmust pass- Keep changes minimal and aligned with existing patterns
License
MIT © CPZ Capital Ltd.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。