Content
# Gravitee AI Workshop
A hands-on workshop to explore **Gravitee Agent Mesh Features** with AI Agents, LLMs, and secure API management using visual inspectors and development tools.
## 🎯 What You'll Learn
Deploy and experiment with:
- **🤖 AI Agent Management**: Secure LLM/Agent exposure with policies
- **🔐 Token Tracking & Guard Rails**: New Gravitee AI security policies
- **📋 Agent Discovery**: Unified agent catalog via A2A Agent Cards
- **🔧 MCP Tools Server**: Make your APIs discoverable as Tools to AI Agents via embedded MCP servers
- **🕵️ Interactive Testing**: Use visual inspectors and tools instead of command-line testing
## 🚀 Quick Start
### 1. Configure Enterprise License
> **⚠️ Important**: This workshop requires a **Gravitee Enterprise License** to access AI-specific policies (Token Tracking, Guard Rails, etc.).
#### Option A: Environment Variable
```bash
export GRAVITEE_LICENSE="PUT_YOUR_BASE64_LICENSE_HERE"
```
#### Option B: Using .env File (Recommended)
The `.env-template` file contains all necessary environment variables with default values.
Rename or copy the `.env-template` to a `.env` file and simply replace `PUT_YOUR_BASE64_LICENSE_HERE` with your actual base64-encoded license key.
### 2. Run the Workshop
```bash
docker compose up -d
```
Wait 2-3 minutes for all services to start and the Ollama model to download.
### 3. Access the Platform
| Service | URL | Description |
|---------|-----|-------------|
| **Gravitee Console** | http://localhost:8084 | API Management Console |
| **Gravitee Portal** | http://localhost:8085 | Developer Portal |
| **A2A Agent** | http://localhost:8080 | AI Agent (A2A Protocol) |
| **Hotel API** | http://localhost:8000 | Demo API with MCP tools |
| **MCP Inspector** | http://localhost:6274 | Visual MCP Protocol Inspector |
| **A2A Inspector** | http://localhost:8005 | Visual A2A Protocol Inspector |
| **Ollama LLM** | http://localhost:11434 | Local LLM Runtime |
### 4. Explore and Test
> **💡 Pro Tip**: Notice how both inspectors use `apim-gateway:8082` URLs instead of direct service URLs. This demonstrates how all AI agent interactions flow through the Gravitee API Gateway for security, monitoring, and policy enforcement!
#### 🕵️ **A2A Inspector** - Agent Protocol Testing
Visit http://localhost:8005 to:
1. **Configure Agent Card URL**:
```
http://apim-gateway:8082/agents/bookings/v1/.well-known/agent-card.json
```
2. **Explore Agent Capabilities**: View agent metadata and available skills
3. **Interactive Chat**: Test conversations with examples like:
- *"List the hotel bookings options please"*
- *"Create a booking for John at Hotel Paris"*
- *"Show me booking details for ID 1"*
4. **Protocol Debugging**: Inspect A2A protocol messages in real-time
#### 🔧 **MCP Inspector** - Tool Protocol Testing
Visit http://localhost:6274 to:
1. **Select Protocol**: Choose **"Streamable HTTP"**
2. **Configure MCP Server URL**:
```
http://apim-gateway:8082/bookings/mcp
```
3. **List Tools**: Discover available booking tools (list, get, create, update, delete)
4. **Call Tools**: Execute tools interactively to:
- Get all bookings
- Create new bookings
- Update existing bookings
- Delete bookings
5. **Protocol Analysis**: Debug MCP tool discovery and execution flow
#### 📬 **Postman Collection** (Coming Soon)
A comprehensive Postman collection will be provided for:
- Complete API testing workflows
- Pre-configured requests for all endpoints
- Example payloads and responses
- Integration testing scenarios
## 🏗️ Workshop Architecture
### **Gravitee API Management Stack**
- **MongoDB**: Data storage for API configurations
- **Elasticsearch**: Analytics and logging storage
- **API Gateway**: Secure API exposure with policies
- **Management API**: Configuration and policy management
- **Console UI**: API management interface
- **Portal UI**: Developer portal for API discovery
### **AI Agent Mesh**
- **A2A Agent**: Hotel booking agent following A2A protocol
- **Ollama LLM**: Local qwen3:0.6b model for natural language processing
- **Hotel API**: Demo API with embedded MCP server
- **MCP Inspector**: Visual MCP protocol debugging and inspection
- **A2A Inspector**: Visual A2A protocol testing and debugging
## 🛡️ Key Features Demonstrated
### **1. Agent Security & Policies**
- **Token Tracking**: Monitor and track AI agent API usage
- **Guard Rails Policy**: Implement safety controls for AI interactions
- **Secure Agent Exposure**: Protect LLMs behind Gravitee gateway
### **2. Agent Discovery & Catalog**
- **A2A Agent Cards**: Standardized agent capability discovery
- **Unified Agent Catalog**: Centralized registry of available agents
- **Capability Mapping**: Understand what each agent can do
### **3. API-to-Agent Integration**
- **MCP Server Embedding**: Make existing APIs discoverable to AI
- **Tool Discovery**: Agents automatically find available API tools
- **OpenAPI Integration**: OAS-described APIs become AI-accessible
## 🧪 Workshop Scenarios
### **Scenario 1: Agent Discovery & Testing**
1. **A2A Inspector**: Open http://localhost:8005
- Configure Agent Card URL: `http://apim-gateway:8082/agents/bookings/v1/.well-known/agent-card.json`
- Explore agent capabilities and metadata
- Test interactive conversations: *"List the hotel bookings options please"*
2. **Gravitee Console**: Browse to http://localhost:8084
- Import the A2A Agent via its Agent Card
- Configure security policies for the agent
### **Scenario 2: Tool Discovery & Integration**
1. **MCP Inspector**: Open http://localhost:6274
- Select "Streamable HTTP" protocol
- Connect to MCP server: `http://apim-gateway:8082/bookings/mcp`
- List available tools and test tool calls interactively
2. **See Integration**: Watch how the A2A Agent uses these tools
- Use A2A Inspector to see real-time tool usage during conversations
- Understand the MCP-to-Agent communication flow through the gateway
### **Scenario 3: Secure Agent Exposure**
1. **Gravitee Management**:
- Create API definition for the A2A Agent
- Apply Token Tracking policy to monitor usage
- Apply Guard Rails policy for safety controls
2. **Test Security**: Use inspectors to verify policy enforcement
- Monitor token usage through Gravitee analytics
- Test guard rail behaviors with various inputs
## 🛑 Stop the Workshop
```bash
docker compose down
```
## 🎓 Learning Outcomes
After completing this workshop, you'll understand:
- ✅ How to **securely expose AI agents** through Gravitee
- ✅ How to implement **AI-specific policies** (Token Tracking, Guard Rails)
- ✅ How to create a **unified agent catalog** using A2A protocol
- ✅ How to make **existing APIs discoverable** to AI agents via MCP
- ✅ How to **integrate LLMs** with enterprise API management
- ✅ How to use **visual inspectors** for protocol debugging and testing
- ✅ How to **interactively test agents** without command-line tools
## 🛠️ Development Tools
This workshop provides modern development and testing tools:
- **Visual Protocol Inspection**: Both A2A and MCP protocols have dedicated visual inspectors
- **Interactive Testing**: No curl commands needed - use browser-based interfaces
- **Real-time Debugging**: See protocol messages and agent behavior in real-time
- **Comprehensive Tooling**: Postman collections and visual inspectors cover all testing scenarios
**Ready to explore the future of AI Agent Management? Let's go! 🚀**
