← Back to Blog

What Is MCP? How AI Agents Use Tools Like File Converters

The Model Context Protocol is quietly changing how AI agents interact with your computer. Here's what it means, why it matters, and what it looks like in practice.

The problem MCP solves

AI assistants like Claude are excellent at understanding your intent ("make this video smaller for Discord") but they can't actually do anything on your computer without tools. Without a standardized way to expose tools, every integration is a custom one-off. Your AI knows you want the file converted, but it has no way to call a converter.

MCP (Model Context Protocol) is an open standard that solves this. It defines how AI agents discover and use external tools over a simple stdio interface. Think of it as a USB port for AI capabilities. Plug in a tool, and any compatible agent can use it.

How MCP works

An MCP server is a program that exposes a set of tools via JSON-RPC over stdio. When an AI agent starts up, it connects to its configured MCP servers and discovers what tools are available. Then, during a conversation, the agent can call those tools when they're relevant.

The flow looks like this:

1. You configure an MCP server in your AI agent's settings
2. The agent connects and discovers available tools
3. During conversation, you ask for something that requires a tool
4. The agent calls the tool with the right parameters
5. The tool runs locally and returns results
6. The agent presents the results to you

All of this happens in the background. From your perspective, you just ask and it's done.

What this looks like with ConvX

ConvX exposes 9 tools via MCP: file conversion, batch conversion, format discovery, preset management, file inspection, and dependency checking. When you add ConvX as an MCP server, your AI agent gains the ability to convert files as naturally as it writes text.

A real conversation

You: "Convert all the screenshots in my Downloads folder to WebP"

Claude: I'll batch convert those for you.
→ calls batch_convert with input_paths, output_format, quality

Claude: Done! 18 files converted. 34MB → 8.2MB (76% smaller).
        Saved to ~/Downloads/converted/

No copy-pasting commands. No alt-tabbing to a terminal. No uploading to a website. You described what you wanted in plain language and it happened locally in seconds.

Why local MCP tools matter

MCP tools run on your machine. This is a critical distinction from cloud-based AI tool integrations. When your AI agent uses ConvX via MCP:

• Your files never leave your computer
• No internet connection is needed for the conversion
• Processing is as fast as your hardware allows, no server queue
• No API rate limits or usage fees per conversion

Which agents support MCP?

As of early 2026, MCP is supported by Claude (desktop and API), Cursor, VS Code (via extensions), and a growing number of other AI tools. The protocol is open, so any agent can add support.

Setting it up

Adding ConvX as an MCP server takes about 30 seconds. Add this to your agent's MCP configuration:

{
  "mcpServers": {
    "convx": {
      "command": "convx",
      "args": ["mcp"]
    }
  }
}

That's it. Restart your agent and it will discover ConvX's 9 tools automatically. See the full MCP guide for detailed setup instructions.

The bigger picture

MCP is still early, but the direction is clear: AI agents are moving from text-only assistants to genuine productivity tools that can take actions on your behalf. File conversion is just one example. The same protocol can expose databases, APIs, file systems, design tools, and anything else an agent might need to help you.

ConvX is one of the first consumer tools to ship with native MCP support. As the ecosystem grows, expect this kind of integration to become the norm, not the exception.