Cognitive

Part 5 · Applied AI Systems

Model Context Protocol

The Model Context Protocol is a shared standard that lets AI applications connect to tools and data in one consistent way.

3 min readBuilds on AI Agents and Tool Use

One protocol connects a host to many tools and data sources.

Key concepts

  • A host is the AI application. A client is its connector. A server provides capabilities.
  • Servers offer three main features: Resources to read, Tools to use, and Prompts as templates.
  • Clients can support Elicitation, which lets a server request more information from the user.
  • MCP reduces the integration problem from M × N custom connections to about M + N implementations.
  • A standard connection is not automatically a safe one. Trust and permissions still matter.

The main participants in the Model Context Protocol.

Figure 26.1. The main participants in the Model Context Protocol.
Each MCP request travels to a server and returns a structured result.

Formal view and evidence

  • Established. MCP uses JSON-RPC 2.0. The 2026-07-28 revision makes requests stateless and self-contained, removes the initialize handshake and protocol sessions, and adds server/discover. Sampling and Roots are deprecated. Tasks moved to an extension.
  • Established. Anthropic introduced MCP in November 2024 and donated it to the Agentic AI Foundation under the Linux Foundation in December 2025.
  • Open problem. Practical trust, identity, and supply-chain controls for large MCP ecosystems.

Common misconception

Common belief

"MCP makes AI integrations secure automatically."

In practice

MCP standardizes the connection. Security depends on what you connect and what you permit.

Key takeaways

  • Hosts use clients to connect to MCP servers.
  • Servers expose tools, resources, and prompts.
  • A standard connection does not make a connected tool trustworthy.

Explore next