CNEquity

What CNEquity does

The CNEquity MCP server provides an MCP read interface for a locally managed financial data lake focused on China markets. Data is stored as Parquet files and can be consumed by Python, SQL, the web console, or an AI agent. The project is aimed at repeated research and historical analysis rather than direct trading or order execution.

The registered data includes securities and trading calendars, daily and optional intraday bars, corporate actions, announcements, financial statements, valuation, ownership, fund flows, margin data, index and industry membership, macro indicators, news, sentiment, regulatory events, futures, options, and derived datasets. The repository describes 55 datasets in its current development tree.

How it works

CNEquity collects data through source adapters and batch workflows, places incoming records in staging, validates them, and publishes curated or derived datasets to the local lake. Incremental runs preserve completed batches, allowing interrupted work to continue without repeating successful downloads. Quality checks and metadata record values such as source, data version, and fetch time.

The MCP server operates over the published local data and is read-only by default. Its query model includes dataset discovery, market-data access, financial-statement access, general dataset queries, code-related parsing, and SQL. Queries can use the project’s research semantics, including adjusted prices, point-in-time financial data, historical universes, and delisted securities, where the relevant data and coverage are available.

Setup and configuration

Install Python 3.10 or newer on macOS, Linux, or Windows, then install the package:

pip install cnequity

Create a demonstration lake with a small recent sample:

cne init --profile demo

For a fuller installation, run cne init, then validate the result with cne check. Daily maintenance uses cne run daily. The initial build and later updates depend on the reachability, history, and publication behavior of public upstream sources; some supplementary sources may require separate credentials.

Start the MCP process with an absolute configuration path:

cne mcp --config /absolute/path/to/cnequity.toml

A Claude Code registration, for example, uses claude mcp add cnequity -- cne mcp --config /absolute/path/to/config.toml. HTTP mode is also available with --http --token; the README notes this mode for clients that connect through a public URL.

Tools and capabilities

The CNEquity MCP server is documented as providing six tool areas:

• Dataset descriptions for discovering available tables and fields.
• Code parsing for working with supported query-related code.
• Market data queries for prices and related market datasets.
• Financial-statement queries for company fundamentals.
• General dataset access across the registered lake.
• SQL queries over published local data.

The underlying query layer supports filters such as symbols and date ranges. Python access can request adjusted prices and strict adjustment checks; point-in-time research can specify an as_of date and strict PIT mode. These semantics are useful when an agent must distinguish information available at a historical decision date from data added later.

Limitations and notes

CNEquity is in the 0.x release series, and the repository documentation may describe a newer implementation than the latest PyPI package. Check the installed CLI version and changelog before upgrading.

The project does not provide a backtesting engine, trading signals, or order placement. It also does not include a prebuilt data lake: users must initialize and populate one. Intraday data, futures, and options are disabled by default and require explicit enablement, with coverage checked against the relevant source.

Freshness does not prove historical completeness or research validity. Strict point-in-time queries may return no rows, and historical-universe queries may refuse to run when evidence is insufficient. The code is licensed under Apache-2.0, while upstream data remains subject to its own terms.