What Is MCP, and Why Is It Useful? How Unravel AI Uses an Open Standard for AI Integration

Model Context Protocol (MCP) is an open standard that gives AI applications and agents a common way to talk to external tools and data sources. It was introduced by Anthropic in November 2024 and has since been adopted broadly across editors, IDEs, and agent frameworks. This post describes what MCP is, how it works, and why we use it at Unravel AI.

What is MCP?

MCP is a protocol that standardizes how an AI client (for example an editor or agent) communicates with a tool server. Before MCP, each integration was a one-off connection built for a single combination of model and API. With MCP, you have one client and many servers, and each server exposes its tools, resources, and prompts over the same interface.

The important part is that MCP is an open standard, not a proprietary product. The original specification and the reference implementations were released by Anthropic, and the protocol has since been developed openly with contributions from the broader ecosystem.

How does it work?

MCP describes a client/server model:

• The MCP client is the application that wants to use tools — an editor like Zed or VS Code, an agent like Khadim, or other software integrating a model.
• The MCP server is a small program that exposes concrete tools, resources, and prompts over MCP’s defined messages.
• The transport is either stdio (the server runs as a local process) or HTTP+SSE (the server runs remotely). In both cases they speak the same protocol.

This means an agent that understands MCP can use any tool that has an MCP server, without a new hand-written integration each time. The tool and the agent are decoupled, and both can evolve independently.

Why is MCP practical?

• Less integration work. One protocol against many servers replaces N×M one-off integrations. A new model can use existing servers; a new server can be used by existing clients.
• Local execution and control. With stdio transport, the server runs as a process on your machine. Tool calls stay local, and you decide which servers run.
• Reusable tools. An MCP server that exposes, say, a database or a filesystem can be written once and used by any MCP-compatible client.
• Context transferred explicitly. The model is told about available tools and resources through the protocol, instead of improvising ad hoc.

How Unravel AI uses MCP

Our agents, Khadim and Blop, use MCP to reach the tools they need. Khadim edits code and runs build and tests locally; MCP servers give it a standardized way to perform file and command operations without binding the agent to one specific environment. Blop drives a browser; MCP lets it expose browser actions as tools over the same interface.

We use MCP because it is an open standard we can read and build on, not because it is a vendor’s closed API. It fits our local-first approach: servers run locally, tool calls are inspectable, and we can swap components without rewriting the agent.

Status

MCP is an existing, working standard with reference implementations, clients, and servers. It was introduced by Anthropic in November 2024 and is since supported by a number of editors, IDEs, and agent frameworks. We follow the specification as it evolves openly.

If you want to understand how Khadim and Blop use MCP in practice, read the Khadim post and the Blop post. Contact us if you want to talk about a concrete integration.

---

This post has been updated to describe MCP accurately: an open standard introduced by Anthropic, not a Microsoft product. We will correct it as the specification and our use of it change.