Hithink Finance Data Service

The Problem It Solves

Financial data querying often poses challenges for developers: users must first understand API interfaces, parameter formats, and specific terms like thscode, increasing integration overhead and error risks. Different data acquisition methods (e.g., CLI tools, MCP services, REST APIs, or Python libraries) are scattered across ecosystems, lacking a unified entry point to simplify interactions. Moreover, handling financial tasks involves multi-step logic: recognizing natural language requirements, disambiguating ticker names, selecting appropriate data pipelines, executing queries, and verifying results—cumbersome and error-prone for engineers.

How the Skill Works

This skill addresses these issues through a unified Agent entry point, with core capabilities and key steps as follows:

• Natural Language Conversion: Users can describe needs in everyday language, such as "Compare Moutai and Ping An Bank trends over the past year." The skill automatically parses this into a clear data task without requiring technical parameters from the user.
• Intelligent Task Routing: Based on user intent, the skill categorizes tasks into types like market data, financials, valuations, indices and sectors, etc. For example, input like "Check current constituents of CSI 300" is routed to an index/sector task, with ticker disambiguation handled (e.g., converting "CSI 300" to a standard index code).
• Access Method Decision: The skill probes the current environment (e.g., whether CLI is installed, if MCP is connected, or a Python project exists) and selects the most suitable primary path:
• CLI: Ideal for automation or integrated local data scenarios, encapsulating remote fetching and local DuckDB.
• MCP: Suited for Chat or IDE sessions, connecting to hosted services.
• REST API: Offers zero-dependency HTTP integration with high flexibility.
• Python SDK: Targeted for research workflows, supporting secondary development.
• Unified Credential Management: All remote methods share a single API Key, automatically found via environment variables (e.g., HITHINK_FINANCE_API_KEY) or user-level credential files, avoiding repeated configuration.
• Execution and Verification: After querying, the skill reports the data source, time range, adjustment criteria (e.g., default forward adjustment forward), result row count, and output path, ensuring results are based on real data (with clear explanations if unavailable).

Use Cases and Limitations

The skill explicitly defines its capability scope to guide appropriate usage:

• Strengths: Primarily supports A-share related data, including market data with adjustment, auction data, financials and indicators, latest valuation snapshots, indices/sectors/feature data (e.g., limit ups, dragon-tiger lists), and public fund information, NAV, and on-exchange quotes.
• Scenarios Requiring Confirmation: When multiple identical tickers cannot be uniquely disambiguated, or when users specify non-default parameters (e.g., specific adjustment methods or output formats), the skill performs brief confirmations.
• Out of Scope: Clearly does not support minute-level K-lines, tick data, Level-2 quotes, nor cover Hong Kong/US stocks, fund subscription/redemption/recommendations, futures/options, or macro data/news announcements/reports. For such requests, it states boundaries directly without providing approximate data.
• Best Practices: Avoid providing full thscode upfront—the skill searches and disambiguates automatically; no need to re-ask for the Key when switching access methods, as unified credentials work cross-method; use minimal bounded requests (e.g., single-stock queries) to verify authentication status, rather than downloading full market data.