Published: 2026-07-03

X's Hosted MCP Server: Live-Data Access for AI Agents Explained

Chapters / key moments (click to jump — plays here on the page)

On June 30, 2026, X launched a hosted MCP (Model Context Protocol) server at api.x.com/mcp that gives AI agents a live, read-only window into X through your own account's OAuth permissions — no custom server to build, no API to hand-wire. It exposes 200+ X API tools (search posts, look up users, read conversations, spot trends, pull bookmarks) and works with clients like Claude Desktop, Cursor, VS Code, and Grok. New to MCP generally? See our MCP tools guide.

Source video

"X AI MCP Server Just Changed AI Agents" by Julian Goldie SEO — Watch on YouTube →

Key Takeaways

What MCP is: the Model Context Protocol is a universal "plug" — instead of custom wiring per app, any AI tool talks to any service the same way. One shape, one socket.
What X shipped (June 30, 2026): a hosted MCP server at api.x.com/mcp. Before this you had to stand up your own server, wire the X API by hand, and keep it running. Now X hosts the bridge — you just connect with your own X account.
A second docs server at docs.x.com/mcp lets your agent read X's own developer documentation on the fly, so it can look up how things work while you build.
Read-only, not a megaphone: the hosted setup is built for reading and pulling data — searching posts, users, conversations, trends, and bookmarks — not for auto-posting on your behalf. Treat it as your agent's live research window into X.
Auth model: connection uses OAuth — you log in with your own X account and grant scoped, revocable permission (never your password). A small open-source helper called xurl runs locally, holds your tokens, and auto-refreshes them so you're not constantly logging back in.
Getting-started flow: (1) create a free X developer app, (2) enable OAuth for it, (3) install and configure xurl, (4) point your AI client at the X MCP server address, (5) enable only the specific tools you need.
The #1 setup mistake: turning on all 200+ tools at once overwhelms the agent and it can't decide which to use. Start with just Search and User Lookup, then add more as you actually need them.
Security hygiene: treat your tokens like private keys — never paste them into a chat or drop them into shared files. And point your agent at the docs server too, not just the main one, to cut back-and-forth.

What you can actually set up from this

Extracted from the video's own transcript — the specifics the original summary left out.

Configuration shown

Endpoints named in the video

Main hosted server: api.x.com/mcp (over 200 X API tools: search posts, look up users, read conversations, spot trends, pull bookmarks). Documentation server: docs.x.com/mcp (lets the assistant read X's developer docs while you build). Announced by the X developer team on June 30, 2026.

Template shown in the video. Adapt to your project — do not copy verbatim without reviewing each line.

Reproducible steps

1. Create an X developer app
Free setup in X's developer area; it gives your tools something to authenticate against.
2. Turn on OAuth for the app
You sign in with your own X account and grant scoped permissions; you do not hand a password to the AI tool, and you can revoke access.
3. Install and set up XRL
A small open-source helper that runs on your machine, stores your login tokens, refreshes them automatically and forwards your AI client's requests to X. (Auto-captioned as "Excel" in places.)
4. Point your AI client at the X MCP server address
The speaker says it works with Cursor, Claude Desktop, VS Code and Grok Build.
5. Choose which tools to expose
Do not switch on all 200-plus. Start with a couple, such as Search and User Lookup, and add more only when needed.

Gotchas

Per what the speaker says X confirmed, the hosted setup is for reading and pulling data, not posting on your behalf.
The most common first-time mistake, per the speaker: enabling every tool at once, which confuses the agent about which to use.
Treat the tokens like private keys: never paste them into a chat or a shared file.
Use the docs server as well as the main one so the assistant can look up how X's API works while it builds.
Promotional video: the speaker is an AI avatar selling a paid community, and shows no screen recording of the setup, so the five steps are described rather than demonstrated. Check each one against X's own developer documentation before relying on it.