# What is MCP? AI agents and DiceyTable > MCP is an open standard that lets an AI app such as Claude use outside tools. DiceyTable's MCP server lets your agent build and test games in your tab. - Question: What is MCP, and why does DiceyTable use it? - Source: https://diceytable.com/docs/ai-agents/concepts/what-is-mcp - Last verified: 2026-10-09 MCP, the Model Context Protocol, is an open standard that lets an AI app use tools made by someone else. DiceyTable offers an MCP server, so an agent like Claude can build, preview and play-test your game in your own browser tab. Think of MCP as one plug shape that fits many sockets. Before it, every AI app needed its own add-on for every service. Now a service builds one MCP server, and any app that speaks MCP can use it. Anthropic first published the standard in 2024. Many AI apps support it today. ## The three parts: app, server and tools MCP has three moving parts. Each one has a plain job. 1. **The MCP app** (often called the client). This is the AI app you chat with: Claude Code, Claude Desktop, Cursor or VS Code with Copilot. 2. **The MCP server.** This is a small program that offers tools. Your MCP app starts it and talks to it. 3. **The tools.** Each tool is a named action with a short description and a list of inputs. One might take a screenshot. Another might add an entity to your scene. When you ask for something, the agent reads the list of tools and picks the right one. The app runs the tool and hands the result back. Then the agent decides what to do next. ![An AI app asks an MCP server for a tool, the server runs it, and the result goes back to the app](https://diceytable.com/assets/build/what-is-mcp--app-server-tools-BOlO50hP.svg "**The basic loop.** The agent picks a tool, the server runs it, and the answer comes back to the chat.") ## What DiceyTable's MCP server is DiceyTable's MCP server is a small program called the bridge. Its npm package is `@diceytable/mcp-bridge`. Your MCP app runs it on your computer, and you never start it by hand. The bridge is a bit unusual. It doesn't hold the tools itself. The tools live in a DiceyTable tab in your browser, and the bridge only carries messages between your agent and that tab. That's why you pair a tab before the agent can do anything. [How the DiceyTable MCP bridge works](https://diceytable.com/docs/ai-agents/concepts/how-it-works) follows one message from start to finish. DiceyTable offers about 150 tools. They cover the editor, the table, the lobby, screenshots and imports. The [tool reference](https://diceytable.com/docs/ai-agents/reference/tools) lists every one. ## Why DiceyTable uses MCP - **You pick the agent.** Any app that speaks MCP works, so you can keep the one you already use. - **The agent works as you.** It uses your tab, your sign-in and your seat. It never gets a second account or a hidden copy of the game. - **You stay in charge.** Nothing happens until you type a pairing code. Big actions, like publishing, still wait for your click. [AI agent safety in DiceyTable](https://diceytable.com/docs/ai-agents/concepts/security-model) covers the rules. ## What MCP doesn't change MCP gives the agent tools. It doesn't give it extra rights. The agent can only do what you could do in the same tab. At a table it sees only what your seat sees. The AI model also stays with your app. DiceyTable doesn't run the model or charge for it. Your AI app's own plan covers that. Agent access in DiceyTable is free on every plan. ## Is MCP the same as WebMCP? No. They're two ways to reach the same tools. - **MCP** connects an AI app on your computer, like Claude Code, through the bridge. - **WebMCP** is a newer browser idea. It lets a web page offer tools to an AI agent built into the browser itself. DiceyTable supports both. A built-in browser agent doesn't use the bridge, but the tab still asks for your consent first. See [in-browser AI agents (WebMCP)](https://diceytable.com/docs/ai-agents/setup/webmcp). ## Ready to try it? The [quick start](https://diceytable.com/docs/ai-agents/quick-start) connects your first agent in about 10 minutes. To see the kind of game you can make this way, visit [Build a board game with AI](https://diceytable.com/features/ai-agents-mcp). ## Related - [Connect an AI agent to DiceyTable](https://diceytable.com/docs/ai-agents/quick-start) - [How the DiceyTable MCP bridge works](https://diceytable.com/docs/ai-agents/concepts/how-it-works) - [AI agent safety in DiceyTable](https://diceytable.com/docs/ai-agents/concepts/security-model)