Model Context Protocol
By Anthropic
Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, for connecting AI assistants and applications to external tools, data sources, and systems through a consistent, model-agnostic interface.
Definition
Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, for connecting AI assistants and applications to external tools, data sources, and systems through a consistent, model-agnostic interface.
Overview
Before MCP, every AI application that wanted to read a file, query a database, or call an external API had to implement a bespoke integration for each data source and each model provider, producing an M×N explosion of custom connectors. MCP addresses this by defining a single client-server protocol: an MCP server exposes a set of "resources" (readable data), "tools" (callable actions), and "prompts" (reusable templates) over a standard JSON-RPC-based interface, and any MCP-compatible client — a chat app, an IDE, an agent framework — can connect to that server without custom glue code. A host application, such as Claude Desktop, Claude Code, or a third-party agent, runs one or more MCP clients, each of which maintains a connection to an MCP server. Servers can run locally over stdio (for example, a server that reads local files or runs shell commands) or remotely over HTTP with server-sent events (for example, a server that wraps a SaaS API like GitHub or Slack). Because the protocol standardizes discovery — clients can query a server for its available tools and their schemas — a model can be handed a new capability simply by connecting it to a new server, without retraining or fine-tuning. MCP has been adopted well beyond Anthropic's own products: OpenAI, Google DeepMind, and a large ecosystem of open-source projects have built MCP clients and servers, and it is often compared to how the Language Server Protocol standardized editor-to-language-tooling communication. For developers, MCP matters because it turns "connect my AI to my data and tools" from a custom engineering project into a matter of picking or writing a conformant server, which is why it underpins much of the current wave of agentic coding tools and enterprise AI assistants.
Specification
- Open, model-agnostic standard for connecting AI applications to tools and data sources
- Client-server architecture with a standardized JSON-RPC message format
- Servers expose resources (data), tools (actions), and prompts (templates)
- Supports both local (stdio) and remote (HTTP/SSE) server transports
- Dynamic capability discovery — clients can query servers for available tools and schemas
- Growing ecosystem of pre-built servers for GitHub, Slack, databases, filesystems, and more
- Adopted across multiple AI vendors, not limited to Anthropic's own products
- Reduces integration effort from custom per-tool connectors to a single protocol
Use Cases
Alternatives
History
The Model Context Protocol (MCP) is an open standard for connecting AI assistants and large language models to the systems where data and tools live — file stores, business applications, and development environments — through a single, standardized interface instead of a bespoke integration per source. It was introduced and open-sourced by Anthropic on November 25, 2024, alongside an initial specification (version 2024-11-05) and SDKs for Python and TypeScript. MCP defines a client-server model in which "servers" expose resources, tools, and prompts that any MCP-compatible "host" application can consume. It was rapidly adopted across the AI industry, including by other major model providers, making it a de facto standard for tool use and context integration.
Sources
- Anthropic — "Introducing the Model Context Protocol" (2024) · as of 2026-07-17
- Model Context Protocol — official documentation · as of 2026-07-17
Frequently Asked Questions
From the Blog
Model Context Protocol (MCP) Explained
The Model Context Protocol is an open standard that lets AI assistants connect to tools and data through one consistent interface, instead of custom integrations.
Read More AI & TechnologyContext Windows Explained: How Much AI Can Read
A context window is the maximum text an AI model can read and reason over at once. Learn how it works, why it matters, and how to work within its limits.
Read More AI & TechnologyWhat Is an AI Context Window and Why It Matters
An AI context window is the maximum amount of text a model can consider at once, measured in tokens. It sets the limits for memory, cost, and accuracy.
Read More AI & TechnologyContextual Retrieval: Adding Document Context to Each Chunk
Contextual retrieval prepends a short, document-aware description to every chunk before embedding and indexing it, restoring the meaning that splitting destroys. This guide covers generating those prefixes cheaply, indexing them in both dense and sparse form, the pitfalls that make them useless, and how to prove they helped on your own corpus.
Read More