What is the Model Context Protocol (MCP)?
The Model Context Protocol (MCP) is an open standard that lets AI applications connect to outside data sources and tools in a consistent way. One connector can then serve many AI assistants and agents instead of being rebuilt for each.
Also known as: MCP
Researched and fact-checked by AI, with no human review. 8 sources listed below. How we verify
Last updated
How it works
Anthropic released MCP as open source on November 25, 2024. The company presented it as one shared protocol to replace the custom integration that each new data source otherwise requires. It credited MCP's creation to David Soria Parra and Justin Spahr-Summers.
The specification defines three roles. A host is the AI application, such as a chat assistant or a coding tool. Clients are connectors inside the host. The architecture document says each client talks to exactly one server. Servers are programs that supply context and capabilities. Messages use JSON-RPC 2.0, an existing format for one program to send requests to another. Servers can offer three kinds of features:
- resources (data for the user or the model)
- prompts (reusable message templates)
- tools (functions the model can ask to have run)
Why it matters
The aim of a shared protocol is that a connector written once can work with many AI products. TechCrunch reported on March 26, 2025, that OpenAI chief executive Sam Altman said the company would add MCP support across its products, starting with its Agents SDK. On December 9, 2025, the Linux Foundation announced that Anthropic was contributing MCP to a new Agentic AI Foundation. It said platinum members at launch included Amazon Web Services, Google, Microsoft and OpenAI. The announcement counted more than 10,000 published MCP servers.
The specification flags security risks. It states that tools amount to arbitrary code execution and that hosts must obtain explicit user consent before invoking one. It adds that the protocol cannot enforce these principles itself, leaving them to the applications that implement it.
Where things stand in 2026
As of October 2026, the project's versioning page lists 2026-07-28 as the current revision. Revisions are identified by date. The maintainers' release post of July 28, 2026, says this revision makes the protocol stateless. It says the opening handshake and session identifiers are gone. Each request carries its own version and capability details, it says, so any server instance can answer it. The specification also lists optional extensions, among them Tasks for long-running operations and MCP Apps for interactive interface elements. The post put downloads of the project's Tier 1 software development kits at close to half a billion a month.
The change list also marks three features, Roots, Sampling and Logging, as deprecated, with a deprecation window of at least twelve months.
Sources
- Specification - Model Context Protocol (revision 2026-07-28), Model Context Protocol project
- Architecture - Model Context Protocol (revision 2026-07-28), Model Context Protocol project
- Key Changes - Model Context Protocol (revision 2026-07-28), Model Context Protocol project
- Versioning - Model Context Protocol, Model Context Protocol project
- The 2026-07-28 Specification, Model Context Protocol Blog
- Introducing the Model Context Protocol, Anthropic
- Linux Foundation Announces the Formation of the Agentic AI Foundation (AAIF), Anchored by New Project Contributions Including Model Context Protocol (MCP), goose and AGENTS.md, The Linux Foundation
- OpenAI adopts rival Anthropic's standard for connecting AI models to data, TechCrunch