What is MCP?

The Model Context Protocol (MCP) is an open standard that enables AI assistants to interact with external applications and data sources in a secure, standardized way.

Core Concepts

Clients and Servers

  • MCP Client: The AI application (Claude Desktop, ChatGPT, etc.) that connects to servers on behalf of the user
  • MCP Server: An application (like a4b.ai) that exposes tools and resources for AI clients to use

Tools

Tools are actions that an AI assistant can invoke. They accept parameters and return results. For example, list_assets is a tool that returns a paginated list of assets.

a4b.ai exposes 23 tools covering assets, workspaces, maintenance tasks, users, invites, and QR code generation.

Resources

Resources are read-only data endpoints that AI clients can browse by URI. They provide context without requiring specific parameters. For example, a4b://stats returns organization-wide statistics.

a4b.ai exposes 6 resource templates via a4b:// URIs.

Transport

MCP supports multiple transport types. a4b.ai uses Streamable HTTP — the server accepts JSON-RPC 2.0 requests over HTTPS at a single endpoint (POST /mcp).

How a4b.ai Uses MCP 

┌──────────────────┐         HTTPS         ┌──────────────┐
│                  │                       │              │
│  AI Assistant    │ ──── JSON-RPC 2.0 ──→ │  a4b.ai      │
│  (Claude, etc.)  │ ←── Tool responses ── │  MCP Server  │
│                  │                       │              │
└──────────────────┘                       └──────────────┘
        │                                          │
        │  1. OAuth 2.1 authentication             │
        │  2. tools/call → execute actions         │
        │  3. resources/read → browse data         │
        │                                          │

The a4b.ai MCP server:

  1. Authenticates via OAuth 2.1 with PKCE (no client secrets needed)
  2. Authorizes every request using role-based access policies
  3. Scopes all data to the authenticated user’s organization (multi-tenant)
  4. Logs every tool call for audit compliance

JSON-RPC 2.0

Under the hood, MCP uses JSON-RPC 2.0 as its message format. You don’t need to write this yourself — your AI client handles it automatically:

// Request: Call a tool
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "list_assets",
    "arguments": { "page": 1 }
  }
}

// Response
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      { "type": "text", "text": "{...}" }
    ]
  }
}

Learn More