X MCP: let an AI agent research and post on X

You want to know what developers said about a launch this week, and all you have is Claude. Claude cannot see X on its own: unless something connects it, the answer comes from what it already knows. That connection is an X MCP server.

What is an X MCP server?

An X MCP server is a program that gives an AI app such as Claude Desktop a set of tools for reaching X. MCP, the Model Context Protocol, is the standard way an AI app calls tools; the server is whatever sits on the other end. There are three kinds, and they differ in what they cost per post and in what they can do.

The first row's prices come from X's pricing page, and the second row from xAI's tool documentation, which says X Search covers "X posts, users, and threads" and bills "token usage" and "tool invocations"; we read both on 11 October 2026. The per-1,000 figures are our arithmetic on those prices.

What does the X API cost for reading?

The X API charges 0.005 USD for each post it returns, so reading 10,000 posts costs 50 USD. X's pricing page states that price, and its changelog dates the start of pay-per-use at 6 February 2026, after a closed pilot announced on 20 October 2025.

Creating a post costs 0.015 USD per request, or 0.200 USD when the post contains a URL. A server built on the API therefore needs credits in a developer account before its first call. The other prices, the monthly cap and what became of the free tier are on the X API pricing page, not repeated here. What 10,000, 100,000 and 1 million reads add up to is worked out in the cost of reading X.

What can an agent do on X through a browser?

An agent in X through a browser MCP works the site the way you do, so it can do whatever your signed-in account can. These are prompts you could give it:

• "Search for browser agent with advanced operators, newest first, and list the ten posts with the most replies."
• "Open this thread and read every reply. What do people ask most?"
• "Check what these 20 accounts posted today and give me one line each."
• "Post this, attach the picture in my folder, and reply to the first three answers."
• "Save my bookmarks to a file, one line per post."

Search was behind a sign-in when we checked on 11 October 2026, and the run further down shows what a logged-out agent gets instead.

How does the agent sign in to X?

The agent signs in to X with a password it never sees and a code it reads from the profile's own inbox. The browser layer below the agent types the password; the agent only names which login to use. If X asks for a code by email, the agent reads it from the mailbox bound to that profile. We describe the four sign-in walls and how a profile answers each in the agent's own X account.

What does an X research run look like?

A research run is one question, a handful of page loads and a summary at the end. We have no X account, so on 11 October 2026 we ran the logged-out version: the question "what did Anthropic, OpenAI and NASA post this week?" and one search, on a headless Scalebrowser profile. It opened each profile three times and the search four times, 13 page loads in about 4 minutes of wall-clock time.

What came back to the reader was one line per account, built from the 15 posts it could see. Anthropic: 5 posts from 7 to 10 October, the newest about a report on unintended model actions seen in evaluations. OpenAI: 5 posts from 30 September to 7 October, the newest, pinned, about GPT-6 and Intelligent UI rolling out in ChatGPT. NASA: 5 posts from 9 October to an hour before the run, the newest a Hubble image of a galaxy cluster.

(opens in a new tab)

The search is where the run stopped being research. All 4 loads ended at a sign-in or an error, so a question that needs search needs the signed-in profile, and we have not measured that part. A question that needs X together with Reddit and LinkedIn is research across Reddit, X and LinkedIn, where a fetch tool read 0 of 3 X posts.

Scalebrowser gives each agent its own isolated browser with a persistent identity, on your own machine, so a run stays signed in, handles the captcha and finishes without anyone watching it.

See how Scalebrowser does it

Is Grok the only AI that can read X?

No. Grok has its own door in xAI's X Search tool, but any agent with a browser can read what its browser can read. Signed in, that is what you read; logged out, it was 5 posts per profile in our run.

Which X MCP fits which job?

A dashboard that pulls millions of posts belongs on the API, and a person who researches or posts from their own account fits the browser. The difference is who has to keep the connection alive and how the cost grows.

If you are still choosing between servers, what an MCP server is covers the parts they all share.

Where does the browser run?

The browser runs on your own machine, so the browser an X run uses keeps your account's sign-in there. For the job pages that go with this one, see the use cases overview and agents that sign in.