Introduction

DeepSeek Harness (DSH) adopts a plugin architecture, aiming to connect various data sources and tools to agents. In enterprise applications, allowing agents to access databases or internal business APIs is a common requirement, but directly exposing credentials or granting unlimited invocation permissions poses risks. The dsh-connect plugin provides a standardized approach for connection and tool registration, addressing security and collaboration issues between data onboarding and agent invocation.

Plugin Purpose

This is the data source connection plugin for DeepSeek Harness, maintained by niuhuoshan. It mainly solves two problems: first, securely connecting enterprise databases and third-party APIs to the DSH environment; second, converting these connections into constrained Agent tools that agents can invoke on demand during conversations.

Core Features

The plugin provides the following core capabilities:
* Database connection and schema discovery: Connects to MySQL, PostgreSQL, and other databases, scans physical structures, and generates metadata that the Agent can retrieve.
* AI semantic modeling: Uses the default DSH model to perform semantic recognition of tables and fields, generating business names, descriptions, and tags.
* Read-only SQL queries: Supports the Agent in generating constrained read-only SQL statements for data queries.
* Fixed HTTP API calls: Configures fixed third-party APIs and registers them as constrained Agent tools; the model cannot modify URLs or parameters.
* Agent tool registration: Automatically registers configured data sources and APIs as tools recognizable by the DSH Agent.
* Metadata synchronization and metric recommendation: Supports incremental synchronization and metric recommendation, and supports export in YAML format.

Installation and Activation

Run the following command in the DSH web profile to install the plugin:

dsh plugin --profile web add @niuhuoshan/dsh-connect

After installation, restart DSH. Open Settings -> DSH Connectors, and you will see the plugin entry. The page contains two tabs: Database and HTTP API.

Typical Usage

Database Integration

  1. In Settings -> DSH Connectors -> Database, click Add Database.
  2. Fill in the connection information (host, port, username, password, etc.) and click Test Connection.
  3. Click Scan to complete physical structure discovery. If semantic recognition is needed, select AI Semantic Enhancement and click AI Modeling.
  4. Ask directly in chat, for example: “What is the daily order volume and sales amount for the East China region this month?” The Agent will retrieve metadata and invoke the query tool.

HTTP API Integration

  1. Add an API in Settings -> DSH Connectors -> HTTP API.
  2. Configure the fixed base URL, path template, authentication method, and parameters.
  3. After the test request succeeds and the API is enabled, describe your requirement in natural language in chat, and the Agent will trigger the call.

Use Cases and Considerations

  • Security boundaries: The database supports read-only operations only; it does not support DDL or multi-statements, and it does not support write operations. HTTP API configuration is fixed; the model cannot modify URLs, methods, or cross-domain redirects.
  • Data masking: Sample data is stored locally only and is not exposed to the Agent or model. Sensitive fields (such as passwords and tokens) are automatically masked.
  • Credential management: Database passwords and API credentials are stored in the DSH credential service and do not appear in list interfaces or Agent return values.
  • Open-source license: The plugin is released under the MIT License.

Summary

dsh-connect provides DSH with a standardized data source integration solution. Through AI semantic modeling and constrained tool invocation, it enables agents to access enterprise data and business interfaces securely and efficiently. For more details, refer to the community directory or the GitHub repository.