Point11

MCP server architecture overview

Understand the host/client/server architecture, transport layers, and core primitives of the Model Context Protocol.

The Model Context Protocol (MCP) is an open protocol that standardizes how AI applications integrate with external data sources, tools, and services. Introduced by Anthropic in November 2024 and donated to the Agentic AI Foundation under the Linux Foundation in December 2025, MCP provides a universal, vendor-neutral interface for connecting AI models to enterprise systems.

Why MCP Matters for Enterprise

Before MCP, every integration between an AI model and an external system required a custom connector. If a brand wanted its product catalog accessible to ChatGPT, Claude, and Gemini, it needed three separate integrations. MCP eliminates this by providing a single protocol that works across all major AI platforms.

As of early 2026, MCP has first-class client support in Claude, ChatGPT, Cursor, VS Code, Gemini, and Microsoft Copilot Studio, with over 10,000 active MCP servers and 97 million monthly SDK downloads.

Architecture: Host, Client, Server

MCP follows a client-server architecture with three participant roles:

  • MCP Host: The AI application (e.g., Claude Desktop, ChatGPT, VS Code) that coordinates one or more MCP clients.
  • MCP Client: A connector within the host that maintains a dedicated 1:1 connection to a single MCP server.
  • MCP Server: A program that provides context, tools, and capabilities to MCP clients. This is what Point11 builds and deploys for enterprise clients.

A single host creates one MCP client per server connection. For example, Claude Desktop might connect to a Point11 product catalog server and a separate analytics server simultaneously, each through its own client.

Two-Layer Architecture

MCP operates on two layers:

Data Layer defines the JSON-RPC 2.0 protocol for communication:

  • Lifecycle management: Connection initialization, capability negotiation, and termination.
  • Core primitives: Tools, Resources, and Prompts exposed by the server.
  • Client primitives: Sampling and Elicitation initiated by the server.
  • Utility features: Progress tracking, cancellation, and notifications.

Transport Layer defines how messages are delivered:

  • stdio: For local servers running on the same machine. Zero network overhead. Ideal for development and local tools.
  • Streamable HTTP: For remote servers (recommended since March 2025). Uses HTTP POST for client-to-server messages and optional Server-Sent Events for streaming. Works with serverless infrastructure such as AWS Lambda.

Core Primitives

MCP servers expose three categories of capabilities:

  • Tools: Executable functions the AI model can invoke. Discovered via tools/list and executed via tools/call. Each tool has a name, description, and JSON Schema input definition.
  • Resources: Read-only data sources providing contextual information. Examples include product catalogs, documentation, and configuration data.
  • Prompts: Reusable templates for structuring LLM interactions. Examples include system prompts, few-shot examples, and workflow templates.

Capability Negotiation

When a client connects, a JSON-RPC initialize handshake occurs. Both sides declare supported capabilities. After initialization, the client sends a notifications/initialized message, and the connection is live.

Servers can dynamically notify clients when available tools change via notifications/tools/list_changed, allowing real-time updates to product catalogs and service endpoints.

The Tasks Primitive (November 2025)

The latest specification introduced Tasks, enabling asynchronous, long-running operations. Any request can become "call now, fetch later" with five lifecycle states: working, input_required, completed, failed, and cancelled. This is critical for enterprise operations like bulk catalog updates, inventory syncs, and complex order processing.

Sources

Deploying your first MCP server

Need help implementing this?

Our team can walk you through the setup.

Talk to Sales