What is the Model Context Protocol (MCP)?
The Model Context Protocol (MCP) is an open standard for connecting AI applications to external tools and data. An application's MCP client connects to MCP servers, which expose tools to call, resources to read and prompts to reuse through one common interface.
What is an MCP server?
An MCP server is a program that exposes a system's capabilities to AI applications in the format MCP defines. A CRM, a database, a file system or an internal API can each sit behind an MCP server, and any MCP-compatible application can then use it without custom integration code.
A server can offer three kinds of things:
| Primitive | What it is | Who decides to use it |
|---|---|---|
| Tools | Functions the model can call, such as "create ticket" or "query orders" | The model |
| Resources | Read-only context, such as a file, a schema or a record | The application |
| Prompts | Reusable templates, often shown as slash commands | The user |
Servers run locally (launched by the app and talking over standard input and output) or remotely over Streamable HTTP, where OAuth is the recommended way to authenticate.
How does the Model Context Protocol work?
Anthropic introduced MCP in November 2024 as an open protocol, and in December 2025 donated it to the Agentic AI Foundation, part of the Linux Foundation. It is now supported by a wide range of AI applications and developer tools. The specification defines three roles:
- Host: the AI application the user works in, such as a coding assistant or desktop chat app.
- Client: a connector inside the host. The host creates one client per server.
- Server: the program that exposes tools, resources and prompts.
Messages use JSON-RPC 2.0. Since the July 2026 revision, every request is self-contained: it carries its own protocol version and capabilities instead of relying on a session handshake. A typical exchange: the client lists the server's tools with tools/list, and the host passes their names, descriptions and input schemas to the model. When the model picks one, the client sends tools/call with the arguments and the host hands the result back to the model.
MCP sits on top of function calling. The model still proposes a structured call; MCP standardizes where the tool lives and how the application reaches it.
What does MCP look like in practice?
A product manager using a desktop AI assistant connects their company's CRM MCP server. They ask: "Which enterprise deals have had no activity in two weeks?" The assistant sees a search_deals tool on the server, calls it with the right filters and summarizes the result. The same server works in a coding assistant or any other MCP host, with no new integration.
MCP connects an agent to tools and data. Agent-to-agent communication is a separate job, handled by protocols such as A2A.
How do SaaS teams use MCP?
Teams use MCP in two directions: to give their own agents access to outside tools, and to expose their product to the AI applications customers already use. On the Foldspace side, a team can connect MCP servers to its Product Agent so the agent can call their tools, and the Docs MCP and Product MCP let the team's coding assistants search the docs and list, read and scaffold the team's AI actions.