MCP (Model Context Protocol)
Quick definition
The Model Context Protocol (MCP) is an open standard that defines how AI applications connect to external tools and data sources. Any compatible assistant can then use any compatible service without a custom integration.
On this page
Why MCP exists
Before MCP, connecting an AI assistant to a tool — a CRM, a file store, a ticketing system — meant writing a specific integration for each pair of assistant and tool. MCP replaces that with one protocol: a service exposes its capabilities once as an MCP server, and every MCP-compatible application can use them.
Anthropic introduced MCP as an open standard in November 2024. In December 2025 it donated the protocol to the Agentic AI Foundation, a fund under the Linux Foundation, so that it is governed in the open. The specification takes inspiration from the Language Server Protocol, which did the same for programming-language support in code editors.
How it works
MCP has three roles. The host is the AI application the user works in. Inside it, a client maintains a connection to each server. A server is a service that offers capabilities. Messages use JSON-RPC 2.0.
Servers can offer three kinds of features: tools, which are functions the model can call (search orders, create a ticket); resources, which are data the user or model can read; and prompts, which are reusable message templates. Remote servers are usually protected with OAuth, so a user grants an assistant access to their account without sharing a password.
Security considerations
The specification is explicit that tools represent arbitrary code execution and must be treated with caution. Hosts should obtain user consent before exposing data to a server or invoking a tool, and descriptions of tool behaviour should be considered untrusted unless they come from a trusted server.
In practice: connect only servers you trust, grant the narrowest scopes possible, and remember that text returned by a tool can contain instructions an attacker hoped the model would follow.
MCP in AI SmartTalk
AI SmartTalk uses MCP in both directions. An assistant can connect to external MCP servers, with OAuth where the server requires it, and use their tools in conversations. And AI SmartTalk is itself a remote MCP server, at https://aismarttalk.tech/api/mcp, protected by OAuth 2.1: from an MCP-compatible application, a user can list their assistants, search and audit their knowledge, add documents or FAQ entries, and review or improve an assistant's instructions — limited to the assistants they are allowed to manage.
Frequently asked questions
Is MCP only for Claude?
No. MCP is an open standard, now under the Linux Foundation's Agentic AI Foundation, and it is supported by many AI applications, development tools and services from different vendors.
What is the difference between MCP and an API?
An API is specific to one service. MCP is a common protocol on top of which services describe their capabilities in a standard way, so an AI application can discover and use them without custom code for each one.
What is an MCP server?
A program or remote service that exposes tools, resources or prompts through the Model Context Protocol, so that AI applications acting as MCP clients can use them.
Is it safe to connect MCP servers to an AI assistant?
It is as safe as the servers you choose and the permissions you grant. Use trusted servers, limit scopes, and review what actions the tools can take.