What the official X MCP server is
The two servers in the official docs
The official X MCP server is a "gateway" that X built to connect AI tools to the X API. MCP (Model Context Protocol) is a common standard for connecting AI tools to external services; through a server that follows it, a compatible AI tool can call a service's features via a defined procedure. For the bigger picture on connecting AI tools to external services, see our guide to the ChatGPT Slack connector.
Two servers (X MCP and Docs MCP)
The official docs describe two servers with different purposes. "X MCP" (api.x.com/mcp) handles X API calls like searching posts and looking up users, while "Docs MCP" (docs.x.com/mcp) searches the developer documentation.
"call X API endpoints (search posts, look up users, bookmarks, trends, news, Articles, and more)" / "Docs MCP (https://docs.x.com/mcp) provides documentation search capabilities with tools like search_x and get_page_x." — from the X MCP docs
You fetch real data with X MCP and look up how to use it with Docs MCP — the roles are split.
Connecting to the X API without your own server
Until now, using the X API from an AI tool meant building, hosting, and authenticating your own MCP server. Because X hosts the server this time, that work goes away. The design lets a user run tools with their own account permissions through a local bridge. (The framing that developers no longer need their own integration comes from reporting such as TechCrunch.)
"The bridge authenticates as you (PKCE flow), so tools act with your account's scopes." — from the X MCP authentication section
Authentication and write limitations
Two authentication paths
The MCP server is powerful, but it can't do everything. Understanding the authentication and the write limits helps you use it safely.
Account-scoped OAuth and a read-only token
There are two authentication paths. The OAuth (PKCE) flow acts as you, so tools operate within your account's scopes. If read-only is enough, you can skip the bridge and connect directly with a static App-only Bearer token.
"The bridge authenticates as you (PKCE flow), so tools act with your account's scopes." / "For read endpoints, you can skip the bridge and point a client straight at the URL with a static App-only Bearer token." — from the X MCP authentication section
Depending on the task, you can use OAuth for write-bearing actions and a token for read-only access.
You can't post (read-centric)
One thing to note on permissions is posting. According to TechCrunch, the MCP tool is not compatible with X's Write API, so you can't have an AI post autonomously. In other words, the expectation is "read-centric" use focused on search and data retrieval. In practice, it's safest to assume it's not suited to proactive posting.
Supported clients and where to use it
Clients with setup instructions
Finally, here is which tools can use it and what tasks it suits.
Use it from Grok, Cursor, Claude Desktop, and more
The official docs provide setup instructions for major MCP-compatible clients, including Grok Build, Cursor, Claude Desktop, and VS Code (GitHub Copilot). If your AI tool supports MCP, you can connect it straight to X's gateway. The benefit is being able to work with X's data from the tools you already use, including Claude.
What it suits
It suits "fetch and analyze" work with X data. Searching posts, gathering trends and news, and looking up specific users can all be done within your AI tool's conversation. Proactive posting, on the other hand, is not expected via MCP for now, so you'll need a separate approach for that.
When passing collected web pages or posts to an AI as long text, converting them to Markdown first keeps their structure intact and tends to improve summary and analysis quality.
