How do AI agents plug into tools, and into each other?

By agreeing on a common standard instead of writing a custom connector for every pair. The Model Context Protocol, or MCP, is an open standard introduced by Anthropic in November 2024 to standardize how AI systems such as language models connect to external tools, systems and data sources. Before it, developers often had to build a custom connector for each data source or tool, which Anthropic called an N-by-M integration problem. The protocol's own documentation compares it to a USB-C port for AI applications.

In practice, an MCP client asks its server for a list of the tools and resources it provides. The server replies with a natural-language description of each tool and the format for calling it, and that description is given to the model. They talk using JSON-RPC 2.0 messages. According to Wikipedia, OpenAI officially adopted MCP in March 2025, and in December 2025 Anthropic donated it to the Agentic AI Foundation, a fund under the Linux Foundation.

A standard is also a new surface for trouble. In April 2025 Invariant Labs described tool poisoning attacks, which hide malicious instructions in a tool's description that the model can read but users cannot see.

The second half of the title is about teams. A multi-agent system is a system of several interacting agents, which can solve problems that are difficult or impossible for a single agent. In one pattern, a central model breaks a task into parts, hands them to worker models and combines the results. In another, one agent proposes and another critiques. Wikipedia lists the problems: few coordination protocols, inconsistent performance, and difficulty debugging.