MCP Extensions are optional capability modules that evolve the Model Context Protocol (MCP) without forcing every implementation to support every feature in the core protocol. They provide a structured way to add behavior such as tasks, apps, or authorization mechanisms.
The extension model matters because MCP is used across local tools, enterprise SaaS systems, IDEs, agent platforms, and gateways. A small core protocol improves interoperability, while extensions let specialized capabilities mature at their own pace.
An MCP extension can define:
- new message types or result shapes;
- additional metadata;
- lifecycle rules;
- authorization behavior;
- UI or task abstractions;
- compatibility requirements; and
- deprecation or migration guidance.
Extensions should be explicitly discoverable and versioned. A client should not assume that a server supports an extension merely because it supports MCP. Conversely, a server should not require an extension unless the client can detect and handle that requirement.
MCP Extensions are related to, but broader than, MCP Authorization Extension. Authorization is one class of extension; other extensions may cover interactive tasks, server-rendered interfaces, or other protocol-level behavior.
For developers building AI agents, extensions create a cleaner boundary between tool access and higher-order experiences. For enterprise operators, they provide a path to standardize governance features without fragmenting the base protocol.
The MCP project described extensions becoming first-class in the 2026-07-28 release candidate, including MCP Apps and Tasks.
The LLM Knowledge Base is a collection of bite-sized explanations for commonly used terms and abbreviations related to Large Language Models and Generative AI.
It's an educational resource that helps you stay up-to-date with the latest developments in AI research and its applications.