AI导哈

市场情报问答后端服务

这是一个面向金融问答与投资研究场景的AI市场情报后端服务，通过RESTful和流式API提供实时股票、加密货币、公司新闻及财务报表等金融数据，并集成MCP金融服务器、Google Search检索与SQLite会话记忆。

  • by akv2011
  • Python
  • 本地部署
  • 更新于
在 GitHub 查看 akv2011/Market_intelligence_MCP

使用说明

Market Intelligence AI Backend Service

AI-powered market intelligence backend service with real-time financial data integration and web search capabilities.

Overview

Backend service providing intelligent market analysis through:

  • MCP (Model Context Protocol) server integration for financial data
  • Google Gemini AI with real-time web search grounding
  • RESTful API endpoints (streaming and non-streaming)
  • Conversational context memory using SQLite
  • Two-stage AI pipeline: ticker extraction and data synthesis

Features

  • MCP Financial Server Integration: Real-time stock prices, crypto data, company news, financial statements
  • HTTP API Endpoints:
    • POST /query - Non-streaming JSON responses
    • POST /query/stream - Server-Sent Events streaming for real-time chat
    • GET /health - Health check endpoint
  • Context Memory: SQLite-based conversation history with session management
  • Google Gemini AI: gemini-3.8-flash by default (set GEMINI_MODEL to change it) with Google Search grounding
  • Errors: provider and tool failures are logged on the server; callers get a fixed message, because provider error text can quote the API key
  • Two-Stage Intelligence:
    • Stage 1: Extract ticker symbols and query type
    • Stage 2: Fetch MCP financial data + Google Search + AI synthesis

MCP server

mcp-server-main/ is an MCP server over the Financial Datasets API, adapted from financial-datasets/mcp-server (MIT, see mcp-server-main/LICENSE). It runs on FastMCP 4 and speaks MCP 2026-07-28 over stdio. The backend starts it and calls it through an MCP client for every query that needs market data.

Tools: get_income_statements, get_balance_sheets, get_cash_flow_statements, get_current_stock_price, get_historical_stock_prices, get_company_news, get_available_crypto_tickers, get_crypto_prices, get_historical_crypto_prices, get_current_crypto_price, get_sec_filings.

It needs a FINANCIAL_DATASETS_API_KEY from financialdatasets.ai. Replace /path/to below with where you cloned this repo.

Claude Code:

claude mcp add market-intel -e FINANCIAL_DATASETS_API_KEY=your_key -- uv --directory /path/to/Market_intelligence_MCP/mcp-server-main run server.py

Codex CLI:

codex mcp add market-intel --env FINANCIAL_DATASETS_API_KEY=your_key -- uv --directory /path/to/Market_intelligence_MCP/mcp-server-main run server.py

Gemini CLI (no -- before the command):

gemini mcp add -s user -e FINANCIAL_DATASETS_API_KEY=your_key market-intel uv --directory /path/to/Market_intelligence_MCP/mcp-server-main run server.py

Claude Desktop (claude_desktop_config.json) and Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "market-intel": {
      "command": "uv",
      "args": ["--directory", "/path/to/Market_intelligence_MCP/mcp-server-main", "run", "server.py"],
      "env": { "FINANCIAL_DATASETS_API_KEY": "your_key" }
    }
  }
}

VS Code: copy .vscode/mcp.json.example to .vscode/mcp.json; it asks for the key on first start.

MCP Inspector:

npx @modelcontextprotocol/inspector uv --directory /path/to/Market_intelligence_MCP/mcp-server-main run server.py

Checked on 2026-10-08: Claude Code 2.1.294 connects and MCP Inspector lists all 11 tools and runs them; Codex CLI 0.156.1 accepts the config; Gemini CLI 0.63.0 ran headless and called get_current_stock_price (which needs a Financial Datasets key with credits to return a price). Gemini CLI marks servers "Disabled" in a folder it does not trust; trust the folder or pass --skip-trust.

Quick Start

1. Install Dependencies

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Configure API Keys

export GEMINI_API_KEY=your_gemini_api_key_here
export FINANCIAL_DATASETS_API_KEY=your_financial_api_key_here

Get your Gemini API key from: https://aistudio.google.com/apikey

3. Start the Server

python -m uvicorn api.main:app --host 0.0.0.0 --port 8000

Usage Examples

Health Check

curl http://localhost:8000/health

Response:

{
  "status": "ok"
}

Non-Streaming Query

Request latest stock market news:

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"message": "Give me the latest news on the stock market."}'

Request stock price:

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"message": "What is NVIDIA stock price?"}'

Request cryptocurrency price:

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"message": "What is Bitcoin price right now?"}'

Response Format:

{
  "session_id": "61933c11-6c76-4a0e-925f-bd0d16e51e34",
  "answer": "The NVIDIA Corporation (NVDA) stock price is $183.16 as of October 10, 2025..."
}

Streaming Query

curl -N -X POST http://localhost:8000/query/stream \
  -H "Content-Type: application/json" \
  -d '{"message": "Give me the latest stock market news"}' \
  --no-buffer

Response Format (Server-Sent Events):

data: Analyzing...

data: Here's the latest stock market news...

event: done
data: ccb2dea9-fb94-4522-ac78-f835000fb7c5

Context Preservation

First query:

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"message": "What is Tesla stock price?"}'

Follow-up query using session_id from previous response:

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{"message": "What about their competitors?", "session_id": "SESSION_ID_FROM_PREVIOUS_RESPONSE"}'

Architecture

Two-Stage AI Pipeline

Stage 1: Ticker Extraction

  • Gemini extracts ticker symbol and query type using JSON mode
  • Determines if financial data is needed

Stage 2: Data Synthesis

  • Calls MCP financial server for real-time data
  • Enables Google Search grounding for latest context
  • Combines all sources and sends to Gemini for synthesis

Data Flow

User Query
    |
Stage 1: Extract Ticker (Gemini)
    |
Stage 2a: Fetch MCP Financial Data
    |
Stage 2b: Google Search Grounding
    |
Stage 2c: Gemini Synthesis
    |
Response with Citations

Project Structure

Market_intelligence_MCP/
├── api/
│   └── main.py              - FastAPI application with endpoints
├── market_intel_service/
│   ├── config.py            - Configuration and API keys
│   ├── llm.py               - Gemini AI client with two-stage pipeline
│   ├── chat.py              - Query orchestration
│   └── memory.py            - SQLite conversation storage
├── mcp-server-main/
│   ├── server.py            - MCP financial data server
│   └── pyproject.toml       - MCP server dependencies
├── tests/
│   └── test_errors.py       - provider errors never reach the caller
├── README.md                - This file
├── requirements.txt         - Python dependencies
└── Dockerfile               - Docker configuration for deployment

Tests

pip install pytest
pytest tests

The tests make the Gemini client fail with an error that quotes an API key and check that neither the full answer nor the streamed one passes it on. Running the server with a wrong key and calling both endpoints shows the same fixed message.

相关 MCP