Devcon MCP Workshop 2026
MCP server integrating Autodesk Platform Services, exposing tools for LLM clients like VS Code Copilot, with OAuth authentication and agentic workflow support.
README
🚀 MCP Server with Autodesk Platform Services (APS)
📌 Overview
This project implements a Model Context Protocol (MCP) server with support for:
- Tool exposure for LLM clients (e.g., VS Code Copilot)
- External API integration
- Autodesk Platform Services (APS) connectivity
- Authentication flows (2-legged & 3-legged OAuth)
- Custom client + agent architecture
It represents a production-oriented agentic AI backend system.
🧠 Architecture
🔷 High-Level Flow
+------------------------+
| LLM Client (VSCode) |
| Copilot / Chat UI |
+-----------+------------+
|
v
+------------------------+
| MCP Server |
| (server.js / aps) |
+-----------+------------+
|
------------------------------------------------
| | |
v v v
+-------------+ +-------------+ +----------------+
| External | | APS APIs | | Custom Logic |
| APIs | | (OAuth) | | / Tools |
+-------------+ +-------------+ +----------------+
🔷 Agent Flow
User Prompt
↓
LLM decides tool
↓
MCP Server executes tool
↓
External API / APS call
↓
Structured response → LLM → User
📁 Project Structure
├── .vscode/
│ └── mcp.json # MCP configuration for VS Code client
│
├── node_modules/ # Installed dependencies
│
├── .env # Environment variables (API keys, secrets)
├── .gitignore # Git ignored files
│
├── agent.js # Agent logic (tool orchestration)
├── aps-server.js # APS integration server (OAuth + APIs)
├── client.js # Custom MCP client implementation
├── server.js # Core MCP server (tool definitions)
│
├── package.json # Project metadata & dependencies
├── package-lock.json # Dependency lock file
⚙️ Tech Stack
- Node.js (ES Modules)
- Model Context Protocol (MCP SDK)
- Zod (schema validation)
- HTTP Streaming Transport
- Autodesk Platform Services (APS)
🔑 Key Components
1. server.js
- Core MCP server
- Registers tools
- Handles client requests
2. aps-server.js
- Integrates Autodesk APIs
- Handles OAuth (2L + 3L)
- Secure API communication
3. agent.js
- Implements agentic behavior
- Decides which tools to call
- Enables multi-step workflows
4. client.js
- Custom MCP client
- Can simulate or replace VS Code client
5. .vscode/mcp.json
- Connects VS Code → MCP server
🚀 Getting Started
1️⃣ Install Dependencies
npm install
2️⃣ Setup Environment
Create .env file:
PORT=3000
# APS Credentials
APS_CLIENT_ID=your_client_id
APS_CLIENT_SECRET=your_client_secret
# External APIs
GEMINI_API_KEY=your_key
3️⃣ Run Server
node server.js
Expected output:
MCP server running at http://localhost:3000/mcp
4️⃣ Configure VS Code
Create:
.vscode/mcp.json
{
"servers": {
"devcon-workshop": {
"type": "http",
"url": "http://localhost:3000/mcp"
},
"devcon-aps": {
"type": "http",
"url": "http://localhost:3001/mcp"
}
}
}
5️⃣ Test Tools
In Copilot Chat:
#add 10 and 20
#greet John in spanish
What's the weather in London?
🧩 Tool Design Pattern
server.registerTool(
"tool_name",
{
description: "Tool description",
inputSchema: {
param: z.string()
}
},
async ({ param }) => ({
content: [{ type: "text", text: "Result" }]
})
);
🔐 Authentication
✅ 2-Legged OAuth
Server-to-server
No user interaction
✅ 3-Legged OAuth
User consent required
Fine-grained permissions
Recommended for production
🏗️ Production Considerations
🔹 Scalability
Stateless architecture
Horizontal scaling possible
🔹 Security
Use .env for secrets
Prefer 3LO for user data
🔹 Reliability
Input validation (Zod)
Structured responses
🔹 Observability Add logging for tool execution
Monitor API failures
🎯 Features
✔ MCP-compliant server
✔ External API integration
✔ APS connectivity
✔ Agent-based execution
✔ Custom client support
🔮 Future Enhancements
Dockerize the application
Deploy on Azure / AWS
Add database-backed tools
Implement caching layer
📜 License
MIT License
🙌 Acknowledgements
-
Autodesk Platform Services (APS)
https://autodesk-platform-services.github.io/mcp-devcon2026/ -
Model Context Protocol (MCP)
-
OpenAI / LLM ecosystem
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。