The project's own example shows one query returning 2201 rows on one site and 100 on the others, with completely different top results, which is exactly why having all three behind one interface beats trusting any single screener. Set a long client timeout: opening a browser and paging through results takes far longer than an API call, and the recommended value in the sample config is 300 seconds.
A Playwright-driven server that types a natural-language screening query into a finance site, intercepts the response, and returns the result table. Three sources are supported, so if one site is down or changes its layout you switch to another.
- Screening queries against Tonghuashun iWenCai, Tongdaxin Wenxiaoda, and Eastmoney's conditional stock screener
- Query types covering stocks, ETFs, funds, indices and sector boards
- Results as a table you can take away in markdown or CSV form
- A second mode that drives web chat providers — Nami Search, Tencent Yuanbao and Baidu AI Search — and returns clean text instead of broken copy-paste
- Three transports: stdio, sse and streamable-http on a port you choose
`pip install mcp_query_table`, plus a Chrome browser it can drive — either an already-running instance reached over the debugging port, or a headless one launched with an explicit user data directory. Log in to any site you have an account with beforehand; there is no automatic login. Keep the browser window wide enough that sites do not switch to their mobile layout, or table extraction fails.
One command — pip install -i https://pypi.org/simple --upgrade mcp_query_table