finance

Finance MCP is an intelligent agent toolkit and MCP server designed for financial research workflows. Built on the FlowLLM framework, it exposes a suite of pre-configured tools tailored for financial data retrieval, analysis, and synthesis, including web searches through DashScope and Tavily, stock data access via Tushare, and live code execution within configurable flows. The server can operate in Stdio mode for local MCP clients or in HTTP/SSE service mode for remote access, allowing you to integrate it into larger research pipelines or dashboards. Users can enable or disable individual flows to tailor the system to their research goals, and the platform supports multi-tool coordination (e.g., DashScope searches combined with historical data analysis) through modular YAML configurations. Typical workflows include price-volume analysis, entity extraction from text, and executable code evaluation within secure sandboxes.

To use the Stdio mode, run the provided configuration with uvx, which communicates over standard input/output. For remote access via HTTP/SSE, start the server with the appropriate environment variables filled (API keys and tokens) and connect via the SSE endpoint or the FastMCP client to programmatically call tools such as dashscope_search, tavily_search, or extract_entities_code. The Quick Start examples in the README illustrate both modes and show how to pass initial model parameters and disable specific flows to match your research scope.

Prerequisites:

• Python 3.10 or newer
• pip (Python package manager)

Install Finance MCP from PyPI:

pip install finance-mcp

(Alternative) Install via uv wrapper for stdio mode if you are using the Stdio workflow:

uv pip install finance-mcp

(Optional) If you plan to run in service mode, prepare environment variables as described in the README (API keys for DashScope, Tavily, Tushare, etc.).

Verify installation by listing installed packages or running a quick help:

python -m finance_mcp --help

Then configure and run as demonstrated in the Quick Start section of the README.

Tips and notes:

• Ensure you provide valid API keys for FLOW_LLM_API_KEY, DASHSCOPE_API_KEY, TUSHARE_API_TOKEN, TAVILY_API_KEY, and BAILIAN_MCP_API_KEY when using service mode.
• In Stdio mode, the model name (llm.default.model_name) and disabled flows can be adjusted to balance cost and capabilities.
• The Finance MCP supports enabling/disabling flows via YAML or the provided config strings; use this to tailor performance and cost.
• When running in service mode, you can connect using the FastMCP Python client for convenient tool discovery and invocation.
• If you encounter rate limits or IP blocking with crawl services, monitor usage and consider enabling higher-level caching as described in the documentation.