Skip to content

llms-install.md

Latest commit

 

History

History
73 lines (49 loc) · 2.64 KB

llms-install.md

File metadata and controls

73 lines (49 loc) · 2.64 KB
Raw

Installing the CNEquity MCP server

Instructions for an AI agent setting up CNEquity for a user. CNEquity is a local China A-share data lake; cne mcp exposes it to MCP clients as six read-only tools. No account or API token is needed.

1. Install

Requires Python 3.10+. Either install the package:

pip install cnequity

or skip installation and let the client run it with uvx cnequity … (needs uv).

2. Create a lake

The server refuses to start without data. Ask the user which they want:

  • Try it in about a minute (recommended first): real data for 5 stocks over the last 30 trading days, in a separate directory.

    cne init --profile demo --data-root ~/cnequity-demo --config-out ~/cnequity-demo/cnequity.toml

  • The full lake: every Shanghai, Shenzhen and Beijing A-share over the last 3 years. The first run can take hours; if it stops, running the same command again resumes it.

    cne init

    This writes configs/cnequity.toml in the current directory.

cne init prints the absolute path of the config it wrote. Use that path below.

3. Register the server

--config must be an absolute path: the client starts the server from a directory of its own choosing, and a relative path resolves to an empty lake.

Cline (cline_mcp_settings.json) and most other clients:

{
  "mcpServers": {
    "cnequity": {
      "command": "cne",
      "args": ["mcp", "--config", "/absolute/path/to/cnequity.toml"],
      "disabled": false,
      "autoApprove": []
    }
  }
}

If cne is not on the PATH the client sees, use the absolute path of the executable (which cne), or "command": "uvx" with "args": ["cnequity", "mcp", "--config", "/absolute/path/to/cnequity.toml"].

ChatGPT and other clients that only accept a remote URL need the HTTP mode and a tunnel; see the MCP guide.

4. Verify

Call the describe_lake tool. It returns the lake's data_root, the datasets it holds and their coverage. If it reports no data, the --config path is wrong or relative.

Tools

Tool Use
describe_lake Coverage, freshness and the query rules; call it first
resolve_symbol Company name or code → 600519.SH-style symbol
query_bars Daily, index and minute bars, optionally adjusted
query_fundamentals Financial statements with a point-in-time as_of cutoff
query_dataset Any other dataset
run_sql One read-only DuckDB SELECT across all datasets

All tools are read-only; ingestion and updates are cne commands (cne run daily).