X just launched its hosted X MCP servers for the X API. Any MCP-compatible AI tool — Grok, Cursor, Claude, VS Code — can now search X’s full post archive, look up users, manage bookmarks, fetch trends, and draft Articles directly, all through your own account’s permissions.
The announcement came from the official X Developers account on June 29, and the documentation is live right now.
Two X MCP Servers, Both Hosted
X is offering two hosted MCP (Model Context Protocol) servers:
X MCP at https://api.x.com/mcp — this is the main API server. It exposes 200+ X API endpoints as callable MCP tools. Your AI agent can search posts, look up any user profile, manage bookmarks and bookmark folders, fetch current trends and news, and even draft and publish Articles.
Docs MCP at https://docs.x.com/mcp — a separate server for documentation. Your AI tool can search and read X API documentation on the fly, without leaving your workflow.
Both are hosted by X, so there is no server to run on your end. The heavy lifting happens on X’s infrastructure.
How It Works
X uses a local bridge called <code>xurl mcp</code> to handle authentication. You do not install anything separately — it runs through npx on first use.
The bridge handles OAuth 2.0 for you. On the first run, it opens your browser for a one-time login to your X account. After that, it caches and auto-refreshes the token forever. All diagnostics go to stderr, and stdout stays a clean JSON-RPC channel for MCP communication.
The requirement is an X developer app with OAuth 2.0 enabled and a registered redirect URI at http://localhost:8080/callback. You pass your CLIENT_ID and CLIENT_SECRET as environment variables, and the bridge handles the rest.
What You Can Do
The X MCP server covers a lot of ground:
- Search the full post archive with X’s search operators
- Look up any user by ID or handle
- Read a user’s posts, timeline, and mentions
- List, add, and remove bookmarks, plus manage bookmark folders
- Get trending topics for any location with a WOEID
- Fetch news stories through X’s news endpoints
- Create draft Articles and publish them
- See likers, reposters, and quoters on any post
That is a genuinely useful set of capabilities for an AI agent. If you are building a Grok agent that needs to search recent posts about a topic, or a Cursor workflow that looks up X users automatically, you no longer need to write custom API integration code.
Setup in Every Major Client
The documentation includes ready-to-use config blocks for every major MCP client:
- Grok Build — add the server with
grok mcp add xapi npx -e CLIENT_ID=... -e CLIENT_SECRET=*** -- -y @xdevplatform/xurl mcp https://api.x.com/mcp - Cursor — add the server block to
~/.cursor/mcp.json. TRT covered Cursor for iOS earlier today — that article also digs into how Cursor works with MCP servers like this one. - Claude Desktop — add to
claude_desktop_config.json - VS Code — add to
.vscode/mcp.json - Any MCP client — universal stdio config:
npx -y @xdevplatform/xurl mcp https://api.x.com/mcp
The first run on any client needs a generous startup timeout (300 seconds recommended) to accommodate the browser login flow. After that, it’s instant.
Docs MCP: Search Docs Without Leaving Your Editor
The Docs MCP server is a simpler companion. It provides two tools: search_x to search across X documentation, and get_page_x to retrieve a specific documentation page. You configure it by adding a URL entry like url: https://docs.x.com/mcp to any MCP client config.
This is useful when you are building with the X API and your AI assistant needs to look up endpoint details or authentication guides without context switching.
What This Means
This is a significant infrastructure move from X. The platform is betting on MCP as the standard protocol for AI agent access, and it is making that access hosted and zero-setup.
Compare this to the pre-MCP world: every AI tool that wanted X data needed a custom API integration, separate auth handling, and individual rate limit management. Now any MCP client gets the full API surface through a standard protocol, with OAuth handled by a shared bridge.
The hosted server model is key here. X is not dropping an OpenAPI spec and asking developers to figure it out. They are running the MCP endpoint themselves, maintaining the protocol compliance, and updating it as the API evolves. For a platform with X’s rate limit complexity and auth requirements, that is the right call.
The Grok Build integration is particularly interesting. X owns Grok, so having native first-class MCP support for the X API from day one gives Grok a capability no other AI assistant has out of the box: real-time access to everything happening on X, searchable and actionable through natural language.
Bottom Line
X launched hosted MCP servers that turn the X API into something any AI agent can use with a one-time login. If you work with MCP-compatible tools, this is one of the most practically useful MCP server launches so far — real-time social data is exactly the kind of thing AI agents need but have been bad at getting.
The authentication is the usual X developer portal friction (create an app, register a redirect URI, copy credentials), but that is a one-time cost. After that, the bridge handles everything silently in the background.
The docs are live, the xurl tool is open source, and the servers are running right now.
