XTapDown

XTapDown MCP Server

An MCP (Model Context Protocol) server that gives any LLM client — Claude Desktop, Cursor, Cline, Continue, or your own agent — direct access to the XTapDown X (Twitter) creator toolkit.

Fourteen tools, read-only, zero auth, free.

Tools (14)

Full coverage of every XTapDown web tool — 4 downloaders (video, GIF, image, full archive) consolidated under a single download_tweet since they share one X syndication endpoint.

Install

Claude Desktop

Add to your claude_desktop_config.json:

{  "mcpServers": {    "xtapdown": {      "command": "npx",      "args": ["-y", "xtapdown-mcp"]    }  }}

Cursor / Cline / Continue

Add a stdio MCP server entry pointing at:

npx -y xtapdown-mcp

From source

git clone https://github.com/farukkolip/xtapdown-mcpcd xtapdown-mcpnpm installnpm run buildnode dist/index.js

Hosted (HTTP)

Set PORT=3000 (or any port) to start the server in HTTP / SSE mode for Smithery-style hosted deployments. Endpoint: POST /mcp and GET /mcp following the MCP Streamable HTTP transport spec.

Example interactions

"Download this tweet for me: https://x.com/.../status/..." → download_tweet returns direct MP4, thumbnail, author handle, like count.

"What's the best time to post on X if my audience is in Türkiye?" → get_best_time_to_post_x returns day-by-day windows in TRT.

"My 50k-follower account got 1,200 likes, 80 reposts, 150 replies. What's my engagement rate?" → calculate_x_engagement_rate returns 2.86% (Good — Mid tier).

"I've been getting 8M impressions a month. Would I qualify for ads revenue share?" → calculate_x_ads_revenue returns: projected 24M / 25M threshold = not yet eligible, ~$60/mo estimated payout if you cross.

Why MCP?

X's API is largely closed to creators. Most third-party tools require paid subscriptions or scrape with hidden rate limits. This server uses X's own public syndication endpoint (the same one embed widgets use) plus inline curated content data, so it works offline-free, no auth, no quota, no surprises.

Once added to your MCP client, the LLM can autonomously fetch tweet data and apply XTapDown's research data to whatever you're working on — content drafts, social audits, strategy threads, monetization math.

Design boundaries

This server is read-only by construction. It has no write tools, no account, no token, no OAuth. Every tool either reads public data through X's syndication endpoint or does local computation. The two tools that produce text (generate_tweet_hook, split_long_text_into_thread) return it to the caller — posting is a separate, human action in a different tool. The approval gate is structural: it is the absence of a post tool, not a flag.

What the model gets back on failure (since 1.3.0):

• Every failed call sets isError: true, so the model can branch on failure instead of pattern-matching an emoji in text.
• Network calls time out after 15 s and report it as a timeout, distinct from a connection failure.
• HTTP 429 from upstream is returned as "rate limited, wait N seconds" (from Retry-After when present), not as a bare status code.
• A protected account (403) and a deleted post (404) are reported as different situations.

What it does not do: it does not add its own rate limit on top of the upstream. A read tool pointed at someone can still be used to bulk-collect public data. The only defence is that it reads what X already serves to anyone.

Data sources

• Tweet data: cdn.syndication.twimg.com/tweet-result (public, auth-free)
• Hashtags / best times / hooks / benchmarks: XTapDown editorial library, 2026 updated
• Engagement math: public follower-based formula (likes + reposts + replies + bookmarks + quotes) / followers × 100
• Ads revenue math: X Premium Creator Ads Revenue Sharing documented terms (25M imp/3mo + 500 verified followers + Premium)

Related

• tiktapdown-mcp — TikTok equivalent
• xtapdown.com — full web toolkit with 17 tools

License

MIT — © Faruk Kolip