Introduction

In the DeepSeek Harness (DSH) plugin ecosystem, generating office documents (such as docx and pptx) is already a highly competitive area. By comparison, plugins targeted at Chinese vertical scenarios, especially those that require localized knowledge support, are relatively scarce. dsh-industry-graph-mcp is designed to address the need for structured data in quantitative research and investment strategy development, such as real-time queries about an individual stock’s industry classification, position in the industrial chain, concept-sector components, and peer competitors. It does not rely on network APIs; instead, it provides factual grounding through a local knowledge graph.

Plugin Overview

This is an MCP server built with Node.js (ESM), focusing on knowledge graph services for A-share industrial chains, Shenwan industries, and concept sectors.

  • Name: dsh-industry-graph-mcp
  • Maintainer: helibeiqi
  • Category: Memory
  • License: MIT
  • Key features: zero dependencies, local-first, no API key required, manually curated seed data

Core Features

This plugin provides 7 tools for querying structured relationships among A-share entities:

  1. industry_of_stock: Query the Shenwan level-1 industry, related concepts, and industrial-chain stage of an individual stock.
    • Parameter: stock (stock code or name)
  2. peers: Query peer competitors in the same industry, with support for excluding the stock itself.
    • Parameters: stock, exclude_self
  3. chain_view: View the upstream and downstream structure of an industrial chain, with the specified stock’s stage highlighted.
    • Parameters: chain (industrial-chain ID or name), stock
  4. concept_members: Query the list of stocks in a given concept sector.
    • Parameter: concept (concept name)
  5. concept_intersect: Compute the intersecting stocks of two concepts, for cross-concept stock selection.
    • Parameters: a, b
  6. stock_search: Fuzzy search by code or name substring to find covered stocks.
    • Parameter: query (search keyword)
  7. graph_stats: View an overview of the current dataset size.
    • Parameters: none

Installation and Enablement

This plugin is provided as a dsh bundle. Before use, merge cordis.patch.yml into your DSH profile.

Insert the following configuration under the plugins.mcp-servers path. Please replace the path in cwd with the actual path on your machine:

- insert:
    - path: plugins.mcp-servers
      items:
        - id: mcp-industry-graph
          name: A股产业链知识图谱
          transport: stdio
          command: !!js process.env.QUANT_MCP_NODE || process.execPath
          args:
            - "C:\\Users\\helib\\dsh-industry-graph-mcp\\industry-graph-mcp-server.mjs"
          cwd: "C:\\Users\\helib\\dsh-industry-graph-mcp"
          enabled: true

The command field uses !!js process.env.QUANT_MCP_NODE || process.execPath, which can prevent path breakage caused by local Node version directory updates (for example, 22.22.2 becoming 22.22.2-2).

Typical Usage

In a DSH conversation, you can directly call the tools above to retrieve structured data.

  • Querying industrial-chain position:
    > “Which industrial chain is CATL in, and what stage? Who are the upstream and downstream players?”
    > → chain_view({chain:"锂电池产业链", stock:"300750"})

  • Querying sector components:
    > “Which stocks are in the baijiu sector?”
    > → concept_members({concept:"白酒"})

  • Cross-concept stock selection:
    > “Which stocks are involved in both semiconductors and CPO?”
    > → concept_intersect({a:"半导体", b:"CPO"})

  • Querying peer competitors:
    > “Who are the peer competitors of LONGi Green Energy?”
    > → peers({stock:"601012"})

Applicable Scenarios and Notes

Applicable scenarios: strategy research, investment research conversations, and quantitative stock selection. When the industry attribute or industrial-chain position of a stock needs to be clarified, using the local knowledge graph instead of model inference can reduce hallucinations.

Important notes:
* Data scale: The current version (v0.1.0) contains manually curated seed data, covering 181 stocks, 31 Shenwan level-1 industries, 52 concept sectors, and 14 industrial chains. The data is designed to be extensible.
* Competitors and positioning: There are already 40–61 A-share data MCP competitors on GitHub, and there are also plugins with a structure similar to this one, such as jinjunnn/tide-plugin. The value of this plugin lies in its local-first approach, zero dependencies, and curated seed data shipped with the repository. It can serve as domain-knowledge assets, but it is not a first-mover placeholder.
* Permission requirements: Because the plugin runs as a local process, ensure that the configured cwd path has read and write permissions.

Conclusion

dsh-industry-graph-mcp gives DSH users the ability to build and maintain an A-share domain knowledge graph locally. Through the standard MCP protocol, it can be seamlessly integrated into the DeepSeek Harness conversation flow, enabling Agents to work with financial entity relationships on the basis of traceable facts.