OpenDirect A2A Agent System

Overview

A comprehensive Node.js implementation of an Agent-to-Agent (A2A) system compliant with the A2A Protocol v0.3.0. The system features dual AI-powered agents (Buyer and Seller) for OpenDirect advertising operations, with integrated Model Context Protocol (MCP) support for tool execution.

Built on a clean modular architecture with 5 independent modules that work together seamlessly.

🎯 Key Features

🤖 Dual AI Agents - Autonomous Buyer and Seller agents with natural language understanding
📋 A2A Protocol v0.3.0 - Full compliance with agent card discovery, JSON-RPC 2.0, and HTTP+JSON transports
🔧 MCP Integration - 33+ auto-generated tools from OpenDirect schemas
🎨 Schema-Driven - Single source of truth from OpenAPI 3.0 schemas
💬 Multi-Provider AI - Support for OpenAI, Anthropic Claude, and Google Gemini
🌐 Web Interface - Ready-to-use test client for agent interactions

🚀 Quick Start (5 Minutes)

Prerequisites

node >= 18.0.0
npm >= 9.0.0

Installation

# Clone repository
git clone https://github.com/InteractiveAdvertisingBureau/mcp-config.git
cd mcp-config

# Install dependencies
npm install

# Install test client
cd client-test && npm install && cd ..

Environment Setup

Create .env file:

NODE_ENV=development
PORT=3000

# AI Provider (choose one)
OPENAI_API_KEY=sk-proj-your-key
OPENAI_MODEL=gpt-4o-mini

# Optional: Other providers
ANTHROPIC_API_KEY=sk-ant-...
GEMINI_API_KEY=...

# MCP Configuration
DEFAULT_ANALYSIS_MODEL=openai
DEFAULT_CHAT_MODEL=openai
MCP_ENABLE_ADMIN_TOOLS=false

Or use the quick-start script:

./quick-start.sh

Start Server

npm start
# Expected: Server on port 3000, MCP integration with 33 tools

Start Test Client (new terminal)

cd client-test && npm start
# Expected: Server on http://localhost:8080

Verify

✅ Agent Card Discovery:

curl http://localhost:3000/a2a/buyer/.well-known/agent-card.json | jq

✅ Web Client:

http://localhost:8080

✅ Send test message:

"create an account for Nike"

🏗️ Architecture Layers

Layer Overview

The system is organized into four distinct layers:

Client Layer - Web UI (client-test/) and A2A-compliant clients communicate via the A2A Protocol v0.3.0 over HTTPS.
A2A Agent Layer - Express.js server (port 3000) hosting Buyer and Seller agents with multiple transport protocols (JSON-RPC 2.0, HTTP+JSON), dynamic agent card generation, and AI-powered execution.
MCP Layer - Schema-driven tool generator using OpenAPI 3.0 specs to dynamically create 33+ tools from the OpenDirect schema, with validation and API handlers via Server-Sent Events (SSE) transport.
Schema Layer - OpenAPI 3.0 specification defining 13 resource types (Account, Order, Line, Product, Creative, etc.), 40+ object definitions, and standardized endpoints with parameters and validations.

Architecture Overview

┌─────────────────────────────────────────────────────────────────────┐
│                         CLIENT LAYER                                 │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐              │
│  │  Web Client  │  │Claude Desktop│  │  Test Client │              │
│  │(client-test) │  │  (MCP HTTP)  │  │    (SDK)     │              │
│  └──────┬───────┘  └──────┬───────┘  └──────┬───────┘              │
└─────────┼──────────────────┼──────────────────┼───────────────────┘
          │                  │                  │
          │ HTTP/JSON        │ SSE              │ HTTP
          │ A2A Protocol     │ MCP Protocol     │ A2A Protocol
          │                  │                  │
┌─────────▼──────────────────▼──────────────────▼───────────────────┐
│                    A2A AGENT LAYER (Port 3000)                      │
│  ┌───────────────────────────────────────────────────────────┐     │
│  │         Client Agent (Orchestrator)                       │     │
│  │  • Autonomous Mode: Auto-selects buyer/seller            │     │
│  │  • Orchestrated Mode: User-directed routing              │     │
│  │  • Session management & conversation history             │     │
│  └─────────────┬────────────────────────┬────────────────────┘     │
│                │                        │                           │
│   ┌────────────▼──────────┐  ┌─────────▼──────────┐               │
│   │   Buyer Agent         │  │   Seller Agent      │               │
│   │  /a2a/buyer           │  │  /a2a/seller        │               │
│   │  ┌─────────────────┐  │  │  ┌─────────────────┐│               │
│   │  │ Agent Card      │  │  │  │ Agent Card      ││               │
│   │  │ - Skills        │  │  │  │ - Skills        ││               │
│   │  │ - Transports    │  │  │  │ - Transports    ││               │
│   │  │ - Examples      │  │  │  │ - Examples      ││               │
│   │  └─────────────────┘  │  │  └─────────────────┘│               │
│   │  Skills:              │  │  Skills:             │               │
│   │  • create_account     │  │  • list_products     │               │
│   │  • create_order       │  │  • search_inventory  │               │
│   │  • search_inventory   │  │  • process_orders    │               │
│   └────────────┬──────────┘  └─────────┬───────────┘               │
└────────────────┼─────────────────────────┼─────────────────────────┘
                 │                         │
                 │ MCP Tool Calls          │ MCP Tool Calls
                 │                         │
┌────────────────▼─────────────────────────▼─────────────────────────┐
│                         MCP LAYER                                    │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐             │
│  │ API Testing  │  │  OpenDirect  │  │Schema-Driven │             │
│  │     MCP      │  │     MCP      │  │     MCP      │             │
│  │  /mcp/sse    │  │/agenticdirect│  │ /schema/mcp  │             │
│  │   (8 tools)  │  │  /mcp/sse    │  │   /sse       │             │
│  │              │  │  (10 tools)  │  │  (33 tools)  │             │
│  │  Transport:  │  │  Transport:  │  │  Transport:  │             │
│  │  SSE         │  │  SSE         │  │  SSE         │             │
│  └──────────────┘  └──────────────┘  └──────┬───────┘             │
└─────────────────────────────────────────────┼───────────────────────┘
                                              │
                                              │ Schema Loading
                                              │ & Validation
                                              │
┌─────────────────────────────────────────────▼───────────────────────┐
│                       SCHEMA LAYER                                   │
│  ┌───────────────────────────────────────────────────────────────┐  │
│  │          OpenDirect OpenAPI 3.0 Schema                        │  │
│  │          (opendirect-mcp-schema.json)                         │  │
│  │                                                               │  │
│  │  📋 13 Resource Types:                                        │  │
│  │     Account, Order, Line, Product, Creative,                 │  │
│  │     Assignment, Organization, ChangeRequest, Message         │  │
│  │                                                               │  │
│  │  🏗️  40+ Object Definitions:                                  │  │
│  │     - Request/Response schemas                               │  │
│  │     - Validation rules (Zod)                                 │  │
│  │     - Parameter specifications                               │  │
│  │                                                               │  │
│  │  ✅ CRUD Operations: create, get, list, update, delete       │  │
│  └───────────────────────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────────────┘

System Components

Client Layer

Web UI (client-test/): Pure HTML/CSS/JavaScript interface
A2A Protocol v0.3.0: Standard agent communication
Features:
- Agent selection (Buyer/Seller)
- Connection management
- Real-time response display
- Message history and session tracking

A2A Agent Layer

Dual Agents: Buyer (/a2a/buyer) and Seller (/a2a/seller)
Agent Card Discovery: /.well-known/agent-card.json
Transports:
- JSON-RPC 2.0 (/jsonrpc)
- HTTP+JSON (/rest)
- MCP tools integration
AI-Powered Execution: OpenAI GPT for natural language understanding
Modes:
- Autonomous: Agent auto-selects sub-agents
- Orchestrated: User-directed routing

MCP Layer

Schema-Driven Tool Generator: Processes OpenAPI 3.0 specs
Validation: Zod schemas for runtime parameter checking
Tool Registry:
- API Testing MCP: 8 tools (register, test, analyze APIs)
- OpenDirect MCP: 10 tools (manual implementations)
- Schema-Driven MCP: 33 tools (auto-generated CRUD)
Transport: Server-Sent Events (SSE) for streaming

Schema Layer

OpenDirect Specification: OpenAPI 3.0 format
13 Resource Types: Account, Order, Line, Product, Creative, etc.
40+ Object Definitions: Complete request/response schemas
Validation Rules: JSON Schema with Zod transformation

A2A Protocol Flow

┌─────────────┐
│   Client    │  1. User sends message: "create account for Nike"
│  (Web UI)   │
└──────┬──────┘
       │
       │ POST /api/a2a/chat
       │ { message: "create account for Nike" }
       │
       ▼
┌─────────────────────────────────────────────────────┐
│           Client Agent (Orchestrator)                │
│                                                      │
│  2. Analyze message with AI (OpenAI GPT)           │
│     → Determine: Account creation needed            │
│     → Select: Buyer Agent                           │
│     → Extract params: { name: "Nike" }             │
└──────┬──────────────────────────────────────────────┘
       │
       │ 3. Route to Buyer Agent
       │
       ▼
┌─────────────────────────────────────────────────────┐
│              Buyer Agent                             │
│             /a2a/buyer                               │
│                                                      │
│  4. Receive message via JSON-RPC 2.0:               │
│     POST /a2a/buyer/jsonrpc                         │
│     {                                                │
│       "method": "sendMessage",                      │
│       "params": {                                    │
│         "message": "create account for Nike"       │
│       }                                              │
│     }                                                │
└──────┬──────────────────────────────────────────────┘
       │
       │ 5. Execute skill: create_account
       │
       ▼
┌─────────────────────────────────────────────────────┐
│               MCP Layer                              │
│        (Schema-Driven MCP Server)                    │
│                                                      │
│  6. Call MCP tool: create_account                   │
│     POST /schema/mcp/sse                            │
│     {                                                │
│       "method": "tools/call",                       │
│       "params": {                                    │
│         "name": "create_account",                   │
│         "arguments": {                              │
│           "name": "Nike",                           │
│           "type": "advertiser"                      │
│         }                                            │
│       }                                              │
│     }                                                │
│                                                      │
│  7. Validate with Zod schema                        │
│  8. Execute account creation logic                  │
└──────┬──────────────────────────────────────────────┘
       │
       │ 9. Return result
       │ { accountId: "acc_123", name: "Nike", ... }
       │
       ▼
┌─────────────────────────────────────────────────────┐
│              Buyer Agent                             │
│                                                      │
│  10. Format response                                 │
│      "✅ Created account for Nike (ID: acc_123)"   │
└──────┬──────────────────────────────────────────────┘
       │
       │ 11. Return to orchestrator
       │
       ▼
┌─────────────────────────────────────────────────────┐
│         Client Agent (Orchestrator)                  │
│                                                      │
│  12. Add to conversation history                     │
│  13. Publish progress events                         │
└──────┬──────────────────────────────────────────────┘
       │
       │ 14. Return to client
       │ {
       │   response: "✅ Created account...",
       │   data: { accountId: "acc_123", ... }
       │ }
       │
       ▼
┌─────────────┐
│   Client    │  15. Display response to user
│  (Web UI)   │
└─────────────┘

Multi-Step Workflow Example:

Message: "Create account for Adidas and create order for Adidas with budget $300"

Step 1: create_account
  └─> Response: { accountId: "acc_456" }
       │
       ▼
Step 2: create_order
  └─> Input: { accountId: "acc_456", budget: 300 }
  └─> Response: { orderId: "ord_789", accountId: "acc_456" }

The orchestrator automatically:

✅ Chains multiple operations
✅ Passes data between steps (account ID)
✅ Maintains conversation context
✅ Handles errors gracefully

Key Components

Component	File/Module	Purpose
Main Server	`server.js`	Express setup, module orchestration, routing
Agent Card	`modules/a2a-protocol/lib/a2aEndpoints.js`	Agent discovery with skills, capabilities, security
Agent Executor	`modules/a2a-protocol/lib/agentExecutor.js`	AI-powered tool selection and task management
Client Agent	`modules/a2a-protocol/lib/clientAgent.js`	Agent orchestrator (autonomous/orchestrated modes)
MCP Server	`modules/schema-driven-mcp/server.js`	Schema-driven tool registry with SSE transport
Tool Generator	`modules/schema-driven-mcp/lib/toolGenerator.js`	Auto-generate CRUD tools from schemas
Schema Loader	`modules/schema-driven-mcp/lib/schemaLoader.js`	Load and validate OpenAPI schemas
Web Client	`client-test/`	HTML/CSS/JS A2A test interface
OpenDirect Schema	`opendirect-mcp-schema.json`	OpenAPI 3.0 resource definitions and validations

📋 Detailed Layer Descriptions

1. Client Layer

Web UI (client-test/): Pure HTML/CSS/JS interface
Direct A2A protocol calls
Agent selection (Buyer/Seller)
Real-time monitoring and status updates
Message streaming support

2. A2A Agent Layer

Express.js server (Port 3000)
Dual agents:
- Buyer: /a2a/buyer
- Seller: /a2a/seller
Agent card discovery: /.well-known/agent-card.json
Skills: Order creation, campaigns, creative management, product search
Multiple transports: JSON-RPC 2.0, HTTP+JSON, MCP tools
AI-powered orchestration: Natural language → tool mapping

3. MCP Layer

Schema-driven tool generator from OpenAPI schemas
Input: OpenDirect OpenAPI 3.0 specification
Output: 33+ executable MCP tools with validation
Transport: Server-Sent Events (SSE) for streaming
Tool categories: CRUD operations, search, validation

4. Schema Layer

OpenDirect JSON specification (OpenAPI 3.0)
13 resource types: Account, Order, Line, Product, Creative, etc.
40+ object definitions with full validation schemas
Parameter validation and response schemas

📂 Project Structure

mcp-config/
├── server.js                       # Main entry point (modular architecture)
├── package.json                    # Dependencies
│
├── modules/                        # 5 Independent modules
│   ├── a2a-protocol/              # ⭐ Agent-to-Agent communication
│   │   ├── server.js              # A2A agent orchestrator
│   │   ├── routes/                # Agent endpoints
│   │   └── lib/                   # Agent cards, executors, SDK integration
│   ├── opendirect-mcp/            # OpenDirect MCP implementation
│   ├── schema-driven-mcp/         # Schema-to-tools generator
│   ├── api-testing-mcp/           # API testing & validation
│   └── ai-chat/                   # Multi-provider AI chat
│
├── shared/                         # Shared utilities
│   ├── config/                    # Centralized configuration
│   ├── database/                  # Database connection
│   ├── utils/                     # Logger, UUID, date utilities
│   └── middleware/                # CORS, error handlers
│
├── client/                         # Frontend clients
│   └── standalone/                # AI chat UI
├── client-test/                    # ⭐ A2A test client
│   ├── index.html                 # Web interface
│   ├── app.js                     # Client logic
│   └── style.css                  # Styling
├── client-test-sdk/                # SDK-based test client
│
├── prompts/                        # AI agent prompts
│   ├── buyer/                     # Buyer agent prompts
│   └── seller/                    # Seller agent prompts
│
├── tests/                          # Test resources
│   ├── schemas/                   # Test schemas
│   └── test.sh                    # Module testing script
│
├── opendirect-mcp-schema.json     # OpenDirect schema
├── claude-desktop-config.json     # MCP client configuration
├── quick-start.sh                 # Quick setup script
├── deploy.sh                      # Cloud deployment script
├── Dockerfile                     # Docker configuration
│
└── a2a-agenticdirect-standalone/  # Standalone A2A module

🤖 A2A Agent System

Agent Card Discovery

Each agent exposes a standard agent card at /.well-known/agent-card.json:

# Buyer Agent Card
curl http://localhost:3000/a2a/buyer/.well-known/agent-card.json

# Seller Agent Card
curl http://localhost:3000/a2a/seller/.well-known/agent-card.json

Agent Card Structure:

Identity: Name, description, version
Skills: Available operations (create_order, search_products, etc.)
Transports: JSON-RPC 2.0, HTTP+JSON, MCP tools
Security: Authentication requirements
Examples: Sample requests and responses

Agent Modes

1. Autonomous Mode (Default)

Agent automatically selects appropriate sub-agents
Natural language understanding
Multi-step workflow orchestration
Example: "Create order for Nike" → Agent calls buyer agent with correct tools

2. Orchestrated Mode

User manually selects target agent
Direct agent communication
More control over agent selection

Agent Communication Endpoints

JSON-RPC 2.0 Transport

POST /a2a/{agent}/jsonrpc
Content-Type: application/json

{
  "jsonrpc": "2.0",
  "method": "sendMessage",
  "params": {
    "message": "create an account for Nike",
    "context": {}
  },
  "id": 1
}

HTTP+JSON Transport

POST /a2a/{agent}/rest
Content-Type: application/json

{
  "message": "search for video ad products",
  "sessionId": "optional-session-id"
}

Client Agent API

# Send message to orchestrator
POST /api/a2a/chat

# Get conversation history
GET /api/a2a/chat/:sessionId/history

# Clear conversation
DELETE /api/a2a/chat/:sessionId

# Get progress events
GET /api/a2a/chat/:sessionId/progress

# Set/get agent mode
POST /api/a2a/mode
GET /api/a2a/mode

🔧 Available Modules

⭐ 1. A2A Protocol (Primary Feature)

Purpose: Agent-to-agent communication using official @a2a-js/sdk with AI orchestration.

Endpoints:

/a2a/agents - Agent discovery
/a2a/buyer/.well-known/agent-card.json - Buyer agent card
/a2a/seller/.well-known/agent-card.json - Seller agent card
/api/a2a/chat - Client agent chat
/api/a2a/mode - Set orchestration mode

Agent Skills:

Buyer Agent:

Create and manage accounts
Create advertising orders
Manage campaigns and budgets
Search for ad inventory
Assign creatives to placements

Seller Agent:

List available products
Process orders
Manage inventory
Handle change requests

Key Features:

✅ Agent card auto-generation with skills, examples, security
✅ Multiple transports: JSON-RPC 2.0, HTTP+JSON, MCP tools
✅ AI-powered natural language understanding (OpenAI GPT)
✅ Autonomous and orchestrated modes
✅ Session management and conversation history
✅ Progress tracking and streaming responses

2. OpenDirect MCP

Purpose: Advertising operations following OpenDirect v2.1 specification.

Endpoints: /agenticdirect/mcp/sse

Tools (10 core operations):

Organization, Account, Order, Line, Creative management
Product search and inventory queries
Change requests and messaging

3. Schema-Driven MCP

Purpose: Auto-generate MCP tools from OpenAPI schemas with runtime validation.

Endpoints: /schema/mcp/sse

Tools (33+ auto-generated):

CRUD operations for all OpenDirect resources
Runtime JSON Schema validation
Hot-reload capability without server restart
Sandbox testing environment

Schema Management API:

POST /api/schema/register      # Register custom schema
POST /api/schema/fetch-url     # Fetch schema from URL
POST /api/schema/reset         # Reset to default
POST /api/schema/test-tool     # Test tool in sandbox
GET  /api/schema/current       # Get current schema
GET  /api/schema/history       # Get registration history

4. API Testing MCP

Purpose: Test and validate REST APIs with AI analysis.

Endpoints: /mcp/sse

Tools: Register APIs, test endpoints, generate AI-powered analysis and test scenarios.

5. AI Chat

Purpose: Multi-provider AI chat with MCP integration.

Endpoints: /chat

Providers: Anthropic Claude, OpenAI GPT, Google Gemini

🧪 Testing

Test Agent Card Discovery

curl http://localhost:3000/a2a/buyer/.well-known/agent-card.json | jq

Expected Response:

{
  "name": "OpenDirect Buyer Agent",
  "version": "1.0.0",
  "skills": [
    {
      "name": "create_account",
      "description": "Create a new buyer account",
      "parameters": { ... }
    }
  ],
  "transports": [
    {
      "type": "jsonrpc",
      "version": "2.0",
      "endpoint": "/a2a/buyer/jsonrpc"
    }
  ]
}

Test JSON-RPC Message

curl -X POST http://localhost:3000/a2a/buyer/jsonrpc \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "sendMessage",
    "params": {
      "message": "create an account for Nike"
    },
    "id": 1
  }' | jq

Test MCP Tools

# List available MCP tools
curl http://localhost:3000/schema/mcp/tools | jq '.tools[].name'

# Test MCP health
curl http://localhost:3000/schema/mcp/health
curl http://localhost:3000/agenticdirect/mcp/health
curl http://localhost:3000/mcp/health

Run All Module Tests

./tests/test.sh

💬 Common Test Scenarios

Buyer Agent Examples

# Account creation
"create an account for Nike"

# Order creation
"create an order for Adidas with budget $50000"

# Inventory search
"search for video ad inventory"

# Campaign setup
"create a campaign for summer sale targeting 25-34 age group"

Seller Agent Examples

# Product listing
"list available products"

# Inventory search
"search for premium ad space"

# Order processing
"process order for account ABC"

# Change requests
"handle change request for order 12345"

🔌 Claude Desktop Integration

HTTP Transport (Recommended)

Create claude-desktop-config.json:

{
  "mcpServers": {
    "api-testing-mcp": {
      "url": "http://localhost:3000/mcp/sse",
      "transport": "http"
    },
    "opendirect-mcp": {
      "url": "http://localhost:3000/agenticdirect/mcp/sse",
      "transport": "http"
    },
    "schema-driven-mcp": {
      "url": "http://localhost:3000/schema/mcp/sse",
      "transport": "http"
    }
  }
}

🐳 Docker Deployment

# Build image
docker build -t mcp-config .

# Run container
docker run -p 3000:3000 \
  -e PORT=3000 \
  -e OPENAI_API_KEY=sk-proj-... \
  -e OPENAI_MODEL=gpt-4o-mini \
  mcp-config

☁️ Cloud Run Deployment

# Deploy to Google Cloud Run
./deploy.sh

# Or manually
gcloud run deploy mcp-config \
  --source . \
  --platform managed \
  --region us-east4 \
  --allow-unauthenticated

🛠️ Troubleshooting

Issue	Solution
Port 3000 in use	`lsof -i :3000` then `kill -9 <PID>`
Missing API key	Verify `.env` contains `OPENAI_API_KEY`
Server not responding	`curl http://localhost:3000/mcp/health`
Client connection fails	Confirm server running, check CORS
Agent card 404	Check `/a2a/buyer/.well-known/agent-card.json`
MCP tools not loading	Verify schema loaded: `GET /api/schema/current`

Debug Mode

# Enable debug logging
NODE_ENV=development node server.js

# Check server logs
tail -f /tmp/server-test.log

# Test each module independently
curl http://localhost:3000/mcp/health
curl http://localhost:3000/agenticdirect/mcp/health
curl http://localhost:3000/schema/mcp/health
curl http://localhost:3000/a2a/agents

📚 API Reference

Health Endpoints

GET /mcp/health                     # API Testing MCP
GET /agenticdirect/mcp/health       # OpenDirect MCP
GET /schema/mcp/health              # Schema-Driven MCP
GET /a2a/agents                     # A2A Protocol
GET /api/ai-health                  # AI Chat

Agent Endpoints

GET  /a2a/agents                                         # List agents
GET  /a2a/{agent}/.well-known/agent-card.json           # Agent card
POST /a2a/{agent}/jsonrpc                               # JSON-RPC 2.0
POST /a2a/{agent}/rest                                  # HTTP+JSON
POST /api/a2a/chat                                      # Client chat
GET  /api/a2a/chat/:sessionId/history                   # History
POST /api/a2a/mode                                      # Set mode

MCP Endpoints

POST /mcp/sse                       # API Testing MCP (SSE)
POST /agenticdirect/mcp/sse         # OpenDirect MCP (SSE)
POST /schema/mcp/sse                # Schema-Driven MCP (SSE)

GET  /schema/mcp/tools              # List generated tools
GET  /schema/mcp/resources          # List resources

✅ Verification Checklist

After setup, verify:

✅ Server running on http://localhost:3000
✅ Test client running on http://localhost:8080
✅ Agent cards accessible:
- http://localhost:3000/a2a/buyer/.well-known/agent-card.json
- http://localhost:3000/a2a/seller/.well-known/agent-card.json
✅ MCP endpoints healthy:
- http://localhost:3000/mcp/health
- http://localhost:3000/schema/mcp/health
✅ Messages sent/received successfully via web client
✅ Response data displayed as JSON
✅ Multi-step workflows execute sequentially

🎨 Architecture Benefits

Protocol Compliance - Full A2A v0.3.0 support with agent cards
Scalability - Modular design allows independent scaling
Extensibility - Schema-driven tool generation from OpenAPI
Maintainability - Clear separation of concerns across modules
AI-Powered - Natural language understanding with multiple AI providers
Interoperability - Standard MCP protocol for tool execution

🚀 Next Steps

Explore Agent Cards - Review buyer and seller capabilities
Test with Web Client - Use http://localhost:8080 for interactive testing
Review MCP Tools - Check /schema/mcp/tools for available operations
Integrate with Claude - Use HTTP transport configuration
Deploy to Production - Use deploy.sh for Cloud Run deployment
Customize Agents - Modify prompts in prompts/buyer/ and prompts/seller/

📚 Resources

A2A Protocol: https://a2a-protocol.org/v0.3.0/specification/
OpenDirect Spec: https://www.iab.com/guidelines/opendirect/
MCP Protocol: https://modelcontextprotocol.io/
JSON-RPC 2.0: https://www.jsonrpc.org/specification
IABTechLab MCP Server: https://mcpclient.iabtechlab.com

Name		Name	Last commit message	Last commit date
Latest commit History 32 Commits
a2a-agenticdirect-standalone		a2a-agenticdirect-standalone
client-test-sdk		client-test-sdk
client-test		client-test
client		client
modules		modules
prompts		prompts
shared		shared
tests		tests
.dockerignore		.dockerignore
.gitignore		.gitignore
Dockerfile		Dockerfile
README.md		README.md
claude-desktop-config.json		claude-desktop-config.json
deploy.sh		deploy.sh
opendirect-mcp-schema.json		opendirect-mcp-schema.json
package-lock.json		package-lock.json
package.json		package.json
quick-start.sh		quick-start.sh
server.js		server.js

Folders and files

Latest commit

History

Repository files navigation

OpenDirect A2A Agent System

Overview

🎯 Key Features

🚀 Quick Start (5 Minutes)

Prerequisites

Installation

Environment Setup

Start Server

Start Test Client (new terminal)

Verify

🏗️ Architecture Layers

Layer Overview

Architecture Overview

System Components

Client Layer

A2A Agent Layer

MCP Layer

Schema Layer

A2A Protocol Flow

Key Components

📋 Detailed Layer Descriptions

1. Client Layer

2. A2A Agent Layer

3. MCP Layer

4. Schema Layer

📂 Project Structure

🤖 A2A Agent System

Agent Card Discovery

Agent Modes

Agent Communication Endpoints

JSON-RPC 2.0 Transport

HTTP+JSON Transport

Client Agent API

🔧 Available Modules

⭐ 1. A2A Protocol (Primary Feature)

2. OpenDirect MCP

3. Schema-Driven MCP

4. API Testing MCP

5. AI Chat

🧪 Testing

Test Agent Card Discovery

Test JSON-RPC Message

Test MCP Tools

Run All Module Tests

💬 Common Test Scenarios

Buyer Agent Examples

Seller Agent Examples

🔌 Claude Desktop Integration

HTTP Transport (Recommended)

🐳 Docker Deployment

☁️ Cloud Run Deployment

🛠️ Troubleshooting

Debug Mode

📚 API Reference

Health Endpoints

Agent Endpoints

MCP Endpoints

✅ Verification Checklist

🎨 Architecture Benefits

🚀 Next Steps

📚 Resources

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages