@memberjunction/a2aserver

MemberJunction implementation of the Google Agent-to-Agent (A2A) protocol server. Exposes MemberJunction AI agents and entity operations as A2A-compliant capabilities, enabling interoperability with other A2A-compatible agents across different platforms and frameworks.

Architecture

graph TD
    subgraph "@memberjunction/a2aserver"
        SRV["A2A Server<br/>Express HTTP"]
        style SRV fill:#2d8659,stroke:#1a5c3a,color:#fff

        CFG["Config<br/>mj.config.cjs"]
        style CFG fill:#b8762f,stroke:#8a5722,color:#fff

        subgraph "Operations"
            AO["AgentOperations<br/>AI Agent Execution"]
            style AO fill:#7c5295,stroke:#563a6b,color:#fff

            EO["EntityOperations<br/>CRUD on Entities"]
            style EO fill:#7c5295,stroke:#563a6b,color:#fff
        end

        subgraph "A2A Protocol"
            AC["Agent Card<br/>/.well-known/agent.json"]
            style AC fill:#2d6a9f,stroke:#1a4971,color:#fff

            TS["Task System<br/>Create, Query, Cancel"]
            style TS fill:#2d6a9f,stroke:#1a4971,color:#fff

            MSG["Messages<br/>User & Agent Parts"]
            style MSG fill:#2d6a9f,stroke:#1a4971,color:#fff

            ART["Artifacts<br/>Structured Output"]
            style ART fill:#2d6a9f,stroke:#1a4971,color:#fff
        end
    end

    SRV --> AO
    SRV --> EO
    SRV --> AC
    SRV --> TS
    TS --> MSG
    TS --> ART

    subgraph "A2A Clients"
        C1["A2A Client Agent"]
        style C1 fill:#b8762f,stroke:#8a5722,color:#fff
        C2["Orchestrator"]
        style C2 fill:#b8762f,stroke:#8a5722,color:#fff
    end

    C1 -->|"HTTP / SSE"| SRV
    C2 -->|"HTTP / SSE"| SRV

    subgraph "MJ Backend"
        AIE["AIEngine"]
        style AIE fill:#2d6a9f,stroke:#1a4971,color:#fff
        AIA["AI Agents"]
        style AIA fill:#2d6a9f,stroke:#1a4971,color:#fff
        MJE["MJ Entities"]
        style MJE fill:#2d6a9f,stroke:#1a4971,color:#fff
    end

    AO --> AIA
    EO --> MJE
    AO --> AIE

About the A2A Protocol

Agent-to-Agent (A2A) is an open protocol developed by Google that enables communication and interoperability between opaque agentic applications. It provides a standardized way for agents built on different platforms to discover, communicate with, and delegate tasks to each other.

• Official A2A Documentation
• A2A Protocol Specification
• GitHub Repository

Installation

npm install @memberjunction/a2aserver

Quick Start

# Start the A2A server (default port 3200)
npx @memberjunction/a2aserver

Key Features

Agent Card Discovery

The server publishes an Agent Card at /.well-known/agent.json following the A2A specification. The card describes the server’s capabilities, supported skills, and authentication requirements.

Task-Based Interaction

All interactions follow the A2A task model:

stateDiagram-v2
    [*] --> pending: Create Task
    pending --> in_progress: Start Processing
    in_progress --> completed: Success
    in_progress --> failed: Error
    in_progress --> cancelled: Cancel Request
    completed --> [*]
    failed --> [*]
    cancelled --> [*]

    classDef stateStyle fill:#2d6a9f,stroke:#1a4971,color:#fff
    class pending,in_progress stateStyle

Entity Operations

CRUD operations on MemberJunction entities exposed through A2A:

Operation	Description
Get	Retrieve a single entity record by primary key
Create	Create a new entity record
Update	Update an existing entity record
Delete	Delete an entity record
Query	Search/filter entity records with RunView

Agent Operations

Execute MemberJunction AI agents through A2A tasks:

• Send messages to agents via A2A task creation
• Receive agent responses as A2A message parts
• Track execution status through task lifecycle
• Access agent artifacts (structured outputs)

Server-Sent Events (SSE)

Streaming support for long-running agent operations. Clients can subscribe to task updates via SSE for real-time status and message delivery.

A2A Protocol Endpoints

Endpoint	Method	Description
/.well-known/agent.json	GET	Agent Card discovery
/tasks/send	POST	Create a new task
/tasks/sendSubscribe	POST	Create task with SSE subscription
/tasks/{id}	GET	Query task status
/tasks/{id}/cancel	POST	Cancel a running task

Configuration

Configure via mj.config.cjs:

module.exports = {
    a2aServerSettings: {
        port: 3200
    },
    databaseSettings: {
        host: 'localhost',
        port: 1433,
        database: 'MemberJunction',
        username: 'sa',
        password: 'password'
    }
};

Required environment variables:

• DB_HOST, DB_PORT, DB_DATABASE, DB_USERNAME, DB_PASSWORD — Database connection
• AI_VENDOR_API_KEY__* — AI provider API keys

Message Format

A2A messages contain typed parts:

// Text part
{ type: 'text', content: 'Analyze sales data for Q3' }

// File part
{ type: 'file', content: '<base64-encoded-data>', metadata: { mimeType: 'application/pdf' } }

// Data part (structured)
{ type: 'data', content: { key: 'value' } }

Dependencies

• @memberjunction/ai-agents — AgentRunner for agent execution
• @memberjunction/ai-core-plus — Extended entity classes
• @memberjunction/aiengine — AIEngine for metadata
• @memberjunction/ai — Core AI abstractions
• @memberjunction/server — MJ server authentication
• @memberjunction/server-bootstrap-lite — Class registration manifest
• @memberjunction/sqlserver-dataprovider — SQL Server data access
• @memberjunction/api-keys — API key management
• @memberjunction/config — Configuration loading
• @memberjunction/encryption — Data encryption
• express — HTTP server framework
• zod — Schema validation