The Model Context Protocol (MCP) is an open protocol for connecting AI applications to external tools, data sources, and reusable prompts. It was introduced by Anthropic and provides a common interface between an MCP host, such as an AI agent or development environment, and one or more MCP servers.
An MCP server can expose three primary capability types:
- Tools: executable operations that can be selected through tool calling.
- Resources: application-controlled data that can be read and included in model context.
- Prompts: reusable templates or workflows offered by the server.
MCP uses a client-server architecture. The host manages the user experience and model interaction, while an MCP client maintains a connection to a specific server. Messages use JSON-RPC, commonly over local standard input/output or Streamable HTTP.
The protocol separates capability discovery from execution. A client can inspect the tools and resources offered by a server before deciding what to make available to the model. This reduces custom integration code and allows the same server to work with multiple compatible hosts.
MCP does not make a server trustworthy. Hosts still need authentication, authorization, user consent, input validation, network restrictions, and controls around consequential actions. Tool descriptions and returned content can influence model behavior, so MCP integrations must also account for indirect prompt injection attacks.
MCP is complementary to the Agent2Agent Protocol (A2A). MCP connects an AI application to tools and context; A2A supports communication and task delegation between independent agents.
The current specification and SDKs are available on the Model Context Protocol website.
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.