
DataSocial
io.github.hashfunction-devv1.0.0Updated Sep 29, 2026
Query a TikTok data warehouse (creators, videos, sounds, daily history) with read-only SQL.
Overview
Lets an assistant write and run read-only SQL against a hosted TikTok data warehouse of creators, videos, sounds and follow edges.
- What it does
- A hosted, stateless MCP server over a ClickHouse warehouse of TikTok data. Tools include list_tables (tables, row counts, what one row is), describe_table (columns, types, sort key), examples (worked questions with runnable SQL and tips), and run_sql, which runs one read-only SELECT returning up to 100 rows. The assistant writes the SQL from a plain question and executes it.
- When to use it
- Useful when you want to explore TikTok creator, video, sound or follow-graph data conversationally without writing SQL by hand. Suited to questions such as fastest-growing large US creators, sounds used in the most new videos, who a handle follows, or average view counts by country and follower band.
- Requirements
- A remote Streamable HTTP endpoint at no local runtime or package is needed. An Authorization header with a bearer key is required, and a free key is obtained by signing in at datasocial.ai. Claude Desktop connects through the mcp-remote npm package, which needs Node.js. Network access to the endpoint is required.
Installation
In SourceWeft
- Open DataSocial in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.
Other MCP clients
Add this to your client's mcpServers config.
{
"mcpServers": {
"datasocial-mcp": {
"type": "http",
"url": "https://api.datasocial.ai/mcp"
}
}
}README
DataSocial MCP: query TikTok data from Claude, Cursor or any MCP client
A hosted MCP server over a ClickHouse warehouse of TikTok data: 4.1B creators, 5.5B videos, 604M sounds and 113.7B follow edges (124.6B rows). Ask your AI assistant a question; it writes the SQL and runs it. Free.
- Endpoint:
https://api.datasocial.ai/mcp(Streamable HTTP, stateless) - Key: free. Sign in at datasocial.ai, open the MCP tab, and copy the command with your key already in it.
- Limits: 100 rows per query, 10 s per query, read-only.
Connect
Claude Code
Cursor (~/.cursor/mcp.json)
Claude Desktop (claude_desktop_config.json, via mcp-remote)
Tools
Things to ask
- "Which US creators with over 1M followers grew fastest this week?"
- "Which sounds are being used in the most new videos today?"
- "Who does @<username> follow?"
- "What's the average view count of videos by creators in Brazil with 100K–500K followers?"
Daily creator and video history started on 2026-09-26; sound history has 106 days since January 2026.
Also
- Plain SQL or plain-English search in the browser, no account: datasocial.ai
- How the data is collected: Scraping TikTok's mobile API at billions of records a day
- The full database without the 100-row limit, or the scraper code: datasocial.ai/buy
This repo holds documentation only; the server is hosted.
Source: README.md at commit f7e89bd
Tools
0Version history
1- v1.0.0LatestSep 29, 2026

