Lava MCP: A-share news heat

Install

Works in

• Claude Code
• Gemini CLI
• SourceCraft
• Koda
• Coddy

npx skillfoxx add mcp/lava-mcp

What it does

The Chinese service Lava (熔岩) counts mentions of stocks and thematic sectors on the A-share market in posts by analysts, institutions and popular finance bloggers. Over MCP an agent gets these counts plus short structured takeaways. Authors appear only as an anonymous agency_id. The server is remote at https://lava.ai-dolphin.com/subscribe/v1/mcp over Streamable HTTP and exposes 18 tools. get_stock_news_heat shows how often a given stock was mentioned and by whom, and get_market_hot_stocks ranks the most discussed stocks over a window from one hour to seven days. get_first_mention_stocks finds stocks nobody mentioned for a week that suddenly come up, and get_accelerating_hot_stocks sorts stocks by growth in attention. For quantitative work there is a daily mention history of up to 92 days (get_stock_heat_panel) and counts of how many distinct authors follow a stock (get_agency_coverage). Other tools score authors on a scale of roughly 450 to 1500 and link news items to stocks without headlines or body text. The repository holds a SKILL.md skill with the call order, a Gemini CLI extension manifest and a local Python proxy that forwards requests to the Lava server.

Who it is for. For traders and quant analysts who follow the Chinese A-share market and want to ask an agent which stocks and sectors are being discussed right now.

Example request

Which A-share stocks were mentioned most in the last 24 hours, and which of them nobody wrote about for the week before?

MCP

Transport

http

Authentication

key is optional

Limitations

Lava is a third-party commercial service with no ties to exchanges or regulators. How it selects authors and computes their scores cannot be verified from public materials, so treat the signals as attention statistics, not investment judgement.

Every tool call costs one request: signup gives a 100 yuan voucher, after that about 5 yuan per 1000 requests, and failed requests are not charged. get_account shows the balance for free. The key starts with lvk_ and goes in the Authorization header; the server also accepts OAuth 2.1 sign-in.

All documentation, the skill and the takeaways in responses are in Chinese. Sector trends and sentiment exist only on the service website or in plans and are not available through the API yet. The sector ranking is capped at a three-day window.

The repository code is only a thin proxy; the data itself lives on and comes from the Lava server. The repository was created in late October 2026. Reachability of the server from Russia was not checked for this card.

How to disable. Remove the server from your MCP client settings; in Claude Code run claude mcp remove lava. If you started the local proxy, stop the python3 proxy/server.py process. You can revoke the key in the Lava dashboard under AI数据服务.

README in short

The repository works as a storefront for the service: tool descriptions with parameters, website screenshots and links to OpenAPI, llms.txt and the MCP Registry entry. Newcomers get three steps: take a key from the dashboard, add the server address to the client and ask in Chinese whether a stock is hot right now.