MCP (Model Context Protocol)
MCP, the Model Context Protocol, is an open standard that lets an AI app connect to outside tools and data through one common interface.
Updated
On this page
What it is
Before MCP, every AI app needed its own plugin format to reach a service. With MCP a service runs one MCP server, and any AI app that speaks the protocol (an MCP client) can list its tools, call them with arguments and read the results.
A server can offer three kinds of things: tools the AI can call, prompts that start a ready-made job, and resources it can read. The client decides when to call a tool, usually after asking you the first time.
In Xodia
Xodia runs an MCP server for your workspace. When you connect Claude Code, the Claude app, Cursor, Codex, VS Code or another MCP app, it gets Xodia's tools to read and edit your site, take screenshots, run checks and save restore points, plus prompts such as build and work, and skills that tell it how to do each job well.
Your AI works on the same live preview you see in Xodia, and every change it makes stays in Xodia for you to review and push. Nothing goes live until you deploy. Connecting an AI works on every plan, Free included, within the plan's limits.
Read more
- The Xodia MCP server reference
- Connect your AI
- Guide: connect an MCP server to build and edit your website
Related terms
- Streamable HTTPStreamable HTTP is the MCP transport for remote servers: the client sends each message as an HTTP POST to one endpoint, and the server can answer with a single response or a stream of events.
- MCP OAuthMCP OAuth is how a remote MCP server signs in the person using an AI app: the app opens the server's sign-in page, you approve access, and the app gets a token it sends with every call.
- llms.txtllms.txt is a plain Markdown file at a site's root, /llms.txt, that tells AI tools what the site is and links to its most useful pages with a line about each.