Model Context Protocol
The Model Context Protocol connects AI models to your tools and data
The Model Context Protocol (MCP) is an open standard for connecting AI models to the tools, data, and systems they need to do useful work. Build an integration once, and any client that speaks the protocol, including Claude, can use it.
What it is
What is the Model Context Protocol?
The Model Context Protocol is an open standard that defines how AI applications talk to external tools and data sources. Instead of writing a separate, custom integration for every model and every system, developers implement the protocol once. A model can then discover what a connected system offers and use it during a task.
The protocol describes a small, consistent vocabulary: tools a model can call to take actions, resources it can read for context, and prompts that package reusable instructions. Because the vocabulary is shared, the same connection works across different applications and different models, rather than being locked to one product.
Tools
Functions a model can call to take an action, such as querying a database, sending a request to an API, or writing a file, within the permissions you set.
Resources
Read-only context a model can pull in, such as documents, records, or files, so its responses are grounded in your real data.
Prompts
Reusable, named instructions a server can offer, so common workflows are packaged once and shared across the applications that connect to it.
Why it matters
Why the Model Context Protocol matters for interoperability
Connecting an AI model to real systems used to mean building bespoke integrations for each pairing of model and tool. That work does not transfer: change the model, or add a new data source, and much of it has to be rebuilt.
An open standard changes the shape of the problem. When a system exposes its capabilities through the Model Context Protocol, any protocol-aware application can use them. A team that builds an MCP server for its internal service makes that service available to the tools it already uses, without rewriting the connection for each one. Because the standard is open, no single vendor controls it, and integrations are not tied to one model or one company.
For developers and technical decision-makers, that means less duplicated integration work, connections that carry across models, and a growing ecosystem of shared servers to build on rather than start from scratch. Interoperability is the point: an open protocol lets the pieces fit together.
Claude and MCP
How Claude uses the Model Context Protocol
Claude acts as an MCP client. When you connect an MCP server to Claude, it can discover the tools and resources that server exposes and use them during a conversation, working within the access you allow.
In practice, that lets Claude read from and act on the systems you connect: pulling context from a document store to answer a question with your own data, or calling a service's API to complete a task you asked for. The same connected server behaves consistently across the Claude apps and the Claude Platform, because both speak the protocol. You decide which servers are connected and what they are permitted to do, so context and capability stay under your control.
Getting started
Getting started with the Model Context Protocol
You can begin using MCP in an afternoon. The fastest path is to connect an existing server to a client that supports the protocol; from there, building your own server is a natural next step.
Read the specification
Start with the MCP documentation and open specification to understand tools, resources, and prompts, and how clients and servers connect.
Connect a server to Claude
Connect an existing MCP server to an MCP-capable client such as Claude, and watch the model discover and use its tools end to end.
Build your own
Use one of the available SDKs to expose your own service as an MCP server, so any protocol-aware client can work with it.
If you are researching how this fits a wider agent stack, see our AI agents hub, and the related guides on agent skills and context engineering.
Start building with MCP and Claude
Read the documentation to understand the standard, then connect a server to Claude and see the Model Context Protocol working with your own tools and data.
Questions
Frequently asked questions about MCP
What is the Model Context Protocol (MCP)?
The Model Context Protocol is an open standard that defines how AI applications connect to external tools and data sources. It gives models a common way to discover and call tools, read resources, and use prompts, so an integration built once can work across any client that supports the protocol.
Who created MCP and is it open?
MCP was introduced by Anthropic as an open standard. The specification and software development kits are openly available, so anyone can build servers and clients that speak the protocol, independent of any single vendor or model.
How does Claude use MCP?
Claude acts as an MCP client. When an MCP server is connected, Claude can discover the tools and resources that server exposes and use them during a conversation, for example to read files, query a database, or call an internal API, within the permissions you set.
What is the difference between an MCP client and an MCP server?
An MCP client is the AI application a person interacts with, such as Claude. An MCP server is a small program that exposes a specific set of tools, resources, or prompts, for example access to a file system or a service's API. Clients connect to one or more servers over the protocol.
How do I get started with MCP?
Read the MCP documentation and specification, then either connect an existing server to an MCP-capable client like Claude, or build your own server using one of the available SDKs. Starting from an existing server is the fastest way to see the protocol working end to end.