XMCP: A Complete Guide to X's Official MCP Server

On June 30, 2026, X released XMCP, its official Model Context Protocol server. It is a significant moment for anyone building AI agents that touch social media. For the first time, X itself is shipping a supported way for AI tools like Claude, Cursor, and other MCP clients to talk to the X platform.

This guide covers what XMCP actually is, what it can do, how to set it up, what it costs, and, just as importantly, where it stops. If you are deciding whether to build directly on XMCP or use a managed layer on top, the tradeoffs are at the end.

What XMCP is and why it matters

XMCP is an MCP server maintained by X. MCP, the Model Context Protocol, is the standard that lets AI assistants call external tools. Before XMCP, connecting an AI agent to X meant either using a community-built MCP server or writing your own wrapper around the X API. Now there is a first-party option.

XMCP ships in two forms:

1. A hosted Streamable HTTP endpoint at https://api.x.com/mcp, so you can point an MCP client at a URL without running anything locally.
2. An open-source repository at github.com/xdevplatform/xmcp that you can self-host and inspect.

There are actually two endpoints inside XMCP. One exposes the X API as MCP tools so agents can act on the platform. The other is a documentation hub that AI tools can query, so an assistant can look up how the X API works before calling it.

Why does this matter beyond convenience? Because X is now legitimizing agents posting to X. For a long time the rules around automation were murky. A first-party MCP server is a clear signal that agent-driven posting is a supported use case, provided you play by X API pricing and terms.

The full documentation lives at docs.x.com/tools/mcp.

What XMCP can do

XMCP is generated directly from the X API OpenAPI spec, which means it exposes more than 140 operations. It is not a read-only or view-only tool. It can write to X in almost every way the API allows.

Highlights of what agents can do through XMCP:

• Publish posts with createPosts
• Like posts with likePost
• Repost with repostPost
• Delete posts with deletePosts
• Send direct messages
• Manage bookmarks
• Create and manage lists
• Mute and block accounts

Because the tool surface is generated from the spec, it tracks the X API closely. If the API supports an operation, XMCP most likely exposes it. That breadth is XMCP's biggest strength: it is the raw platform, wrapped for agents.

Setup walkthrough

Setting up XMCP has three parts: get credentials, then connect a client.

1. Get an X developer account

XMCP authenticates using X API credentials, so the first step is an X developer account. If you do not already have one, sign up through the X developer portal and create an app. This is the same friction any direct X API integration involves, and it is a real step, not a formality. Approval and app configuration take time.

2. Get your credentials

Inside your app, generate the credentials XMCP needs. XMCP supports OAuth 1.0a and OAuth 2.0. In practice that means you will use either a consumer key and secret, or a bearer token, depending on the operations you need. Write actions on behalf of a user generally require user-context OAuth, while some read operations work with an app-only bearer token.

3. Connect from Claude, Cursor, or another MCP client

For the hosted endpoint, add https://api.x.com/mcp as an MCP server in your client of choice and supply your credentials. Most MCP clients accept a remote server URL plus an authorization header. For self-hosting, clone the XMCP repo, run it with your credentials as environment variables, and point your client at your local instance.

Once connected, your AI assistant will see the XMCP tools and can call operations like createPosts directly from a conversation. The exact configuration format depends on your client, and the official docs have per-client instructions.

What XMCP costs

XMCP itself has no monthly subscription. You are not paying for the MCP server. You are paying for the X API usage that flows through it.

That means XMCP costs whatever your X API consumption costs, on a pay-per-use basis against standard X API pricing. Reads and writes count against your plan the same way they would if you called the API directly.

One hard limit to know about: X applies a cap of 2 million post reads per month before it requires an Enterprise agreement. If your agent reads at high volume, you can hit that ceiling and need to move to an Enterprise contract. For most single-account posting workflows this is not a concern, but for anything that reads timelines or search results at scale, it is a real number to plan around.

Where XMCP stops

XMCP is a faithful wrapper over the X API, which means it inherits the API's limits. Three matter most.

No scheduling. The X API has no scheduled-posts endpoint, so XMCP has no scheduling tool. Every createPosts call publishes immediately. If you want a post to go out at 9am tomorrow, XMCP cannot hold it for you. You would need to keep a process running and fire the call at exactly the right moment yourself.

No thread composer. There is no single tool that takes a list of posts and chains them into a thread with correct reply references. You can build a thread by posting each part and linking replies manually, but XMCP does not do the orchestration for you.

No evergreen recycling and no aggregated analytics. XMCP gives you the raw operations. It does not recycle your best posts on a cooldown, and it does not roll individual metrics into dashboards. Those are workflow features that live above the API, not inside it.

There is also the developer-account requirement itself, which is friction for anyone who just wants their agent to post and does not want to manage OAuth apps and API credentials.

When XMCP is the right choice

XMCP is the right tool when you want the raw X platform surface and are comfortable operating at the API level. Reach for it when:

• You need operations beyond publishing, like direct messages, lists, mutes, blocks, or moderation actions. This is where XMCP shines, because few managed tools expose the full API.
• You already have an X developer account and API credentials, so the setup friction is behind you.
• You are building something that maps cleanly to individual API calls and you do not need scheduling, threads, or analytics on top.
• You want a first-party, open-source integration you can inspect and self-host.

When a managed MCP layer makes more sense

If your goal is a publishing workflow rather than raw API access, a managed MCP layer usually fits better. Scheduling, thread composition, evergreen recycling, and aggregated analytics are exactly the features that do not exist in the X API and therefore cannot exist in XMCP. A managed server adds them, and typically handles X authentication so you skip the developer-account step.

For transparency, this is what OpenTweet does. Its hosted MCP server at https://mcp.opentweet.io/mcp uses the same Streamable HTTP transport as XMCP, authenticates with a simple API key (Authorization: Bearer ot_...), and exposes 30 tools including scheduling, threads, an evergreen queue, analytics, and multi-account support. It runs on flat pricing from $11.99 per month, and it does not require an X developer account. You can see a direct feature-by-feature breakdown on the XMCP comparison page, and setup instructions on the Twitter MCP server page.

The honest framing: XMCP gives you everything the X API can do and nothing it cannot. A managed layer trades some raw API breadth for the publishing workflow features that most content and marketing use cases actually need.

Honest comparison

Capability	XMCP (official)	Managed MCP layer
Post, like, repost, delete	Yes	Yes
DMs, lists, mute/block	Yes	Usually limited
Full X API surface (140+ ops)	Yes	No, curated subset
Scheduling	No	Yes
Thread composer	No	Yes
Evergreen recycling	No	Yes
Aggregated analytics	No	Yes
X developer account needed	Yes	Usually no
Pricing model	Pay-per-use (X API)	Often flat subscription
Open source / self-host	Yes	Varies

Neither is strictly better. XMCP is the platform; a managed layer is a workflow. Pick based on whether you need the raw API surface or a publishing pipeline.

FAQ

Can XMCP post tweets?
Yes. XMCP exposes the createPosts operation, so a connected AI agent can publish posts directly to X. It can also like, repost, delete, send DMs, and manage bookmarks and lists. Every post publishes immediately when the tool is called.

Can XMCP schedule tweets?
No. XMCP has no scheduling capability because the X API has no scheduled-posts endpoint. Every call publishes right away. To queue posts for later you need a separate layer that stores the post and publishes it on a timer.

What does XMCP cost?
XMCP has no subscription of its own. You pay per use against standard X API pricing, based on the operations and read volume you consume. X caps usage at 2 million post reads per month before requiring an Enterprise agreement.

Do I need an X developer account for XMCP?
Yes. XMCP authenticates with X API credentials, so you need a developer account and an app with OAuth 1.0a or OAuth 2.0 credentials, or a bearer token.

What is the difference between XMCP and a managed MCP server?
XMCP is a thin wrapper over the raw X API and inherits its limits, including no scheduling and the developer-account requirement. A managed MCP server adds a publishing layer with scheduling, threads, and analytics, and usually handles X authentication for you.

Is XMCP open source?
Yes. The repository is public at github.com/xdevplatform/xmcp, so you can inspect the tools or self-host the server instead of using the hosted endpoint.

Bottom line

XMCP is a genuine step forward. X shipping a first-party MCP server means agent-driven posting is now a supported, documented path, with the full X API behind it. If you need the raw platform, including DMs, lists, and moderation, and you already have developer credentials, XMCP is the natural choice.

If you need scheduling, thread composition, evergreen recycling, or analytics, those features live above the X API, so you will want a managed layer such as OpenTweet's MCP server on top. Start from what your workflow actually requires, then pick the tool that matches it.