Download src/agent.py from margotbtn/finsight: direct link, hf CLI and curl.
- Browser
- Download file 7.51 kB
-
https://huggingface.co/spaces/margotbtn/finsight/resolve/main/src/agent.py
- Command line
-
hf download hf://spaces/margotbtn/finsight/src/agent.py
-
curl -L -o agent.py https://huggingface.co/spaces/margotbtn/finsight/resolve/main/src/agent.py
7.51 kB
| #!/usr/bin/env python3 | |
| """ | |
| Market Intelligence Agent | |
| Answers investment questions by orchestrating two MCP servers: | |
| - mcp_sql/server.py → structured price, indicator, fundamental data (DuckDB) | |
| - mcp_rag/server.py → semantic news search (ChromaDB) | |
| Usage: | |
| python agent.py "Is NVIDIA a good buy right now?" | |
| python agent.py # interactive mode | |
| """ | |
| import asyncio | |
| import json | |
| import os | |
| import sys | |
| from collections.abc import AsyncGenerator | |
| import anthropic | |
| from dotenv import load_dotenv | |
| from mcp import ClientSession, StdioServerParameters | |
| from mcp.client.stdio import stdio_client | |
| load_dotenv() | |
| ROOT_DIR = os.path.dirname(os.path.abspath(__file__)) | |
| SYSTEM_PROMPT = """You are a financial market intelligence analyst with access to real market data. | |
| You have two data sources: | |
| 1. **SQL database** — stock prices (OHLCV), technical indicators (RSI-14, MACD, SMA-20/50/200), and fundamentals (P/E, EPS, market cap, sector) for 25 major tickers. | |
| 2. **News database** — recent financial news articles indexed for semantic search. | |
| When answering an investment question, always follow this process: | |
| 1. Call `get_indicators` for each relevant ticker to retrieve technicals and fundamentals. | |
| 2. Call `search_news` with a descriptive query and the ticker to surface recent events and sentiment. | |
| 3. If the question compares multiple stocks, call `compare_stocks` as well. | |
| 4. Synthesise all signals into a structured answer. | |
| Structure your response with these sections: | |
| - **Technical picture**: RSI level, MACD trend, price vs key SMAs, golden cross/death cross | |
| - **Fundamental context**: valuation (P/E), market cap, sector, EPS | |
| - **News & sentiment**: key recent headlines and what they imply | |
| - **Overall assessment**: balanced conclusion based on the data | |
| Always ground your claims in specific numbers from the tools. Your output is rendered as markdown: escape every dollar sign as \\$ to prevent LaTeX interpretation (e.g. write \\$150.00 not $150.00). End every response with: | |
| "*Disclaimer: This is a data-driven analysis tool, not licensed financial advice. Always do your own research before investing.*" | |
| """ | |
| # --------------------------------------------------------------------------- | |
| # Core: async generator — usable from both CLI and Streamlit | |
| # --------------------------------------------------------------------------- | |
| async def run_agent_stream( | |
| question: str, api_key: str | |
| ) -> AsyncGenerator[dict, None]: | |
| """ | |
| Async generator that yields typed events: | |
| {"type": "log", "text": "..."} — tool call activity | |
| {"type": "result", "text": "..."} — final markdown answer | |
| """ | |
| sql_params = StdioServerParameters( | |
| command=sys.executable, | |
| args=[os.path.join(ROOT_DIR, "mcp_sql", "server.py")], | |
| ) | |
| rag_params = StdioServerParameters( | |
| command=sys.executable, | |
| args=[os.path.join(ROOT_DIR, "mcp_rag", "server.py")], | |
| ) | |
| async with stdio_client(sql_params) as (sql_r, sql_w): | |
| async with ClientSession(sql_r, sql_w) as sql_session: | |
| await sql_session.initialize() | |
| async with stdio_client(rag_params) as (rag_r, rag_w): | |
| async with ClientSession(rag_r, rag_w) as rag_session: | |
| await rag_session.initialize() | |
| sql_tools = (await sql_session.list_tools()).tools | |
| rag_tools = (await rag_session.list_tools()).tools | |
| tool_sessions: dict[str, ClientSession] = { | |
| t.name: sql_session for t in sql_tools | |
| } | { | |
| t.name: rag_session for t in rag_tools | |
| } | |
| anthropic_tools = [ | |
| { | |
| "name": t.name, | |
| "description": t.description or "", | |
| "input_schema": t.inputSchema, | |
| } | |
| for t in (*sql_tools, *rag_tools) | |
| ] | |
| client = anthropic.Anthropic(api_key=api_key) | |
| messages: list[dict] = [{"role": "user", "content": question}] | |
| yield {"type": "log", "text": "Analysing..."} | |
| while True: | |
| response = client.messages.create( | |
| model="claude-sonnet-4-6", | |
| max_tokens=4096, | |
| system=SYSTEM_PROMPT, | |
| tools=anthropic_tools, | |
| messages=messages, | |
| ) | |
| tool_uses = [b for b in response.content if b.type == "tool_use"] | |
| if tool_uses: | |
| names = ", ".join(b.name for b in tool_uses) | |
| yield {"type": "log", "text": f"Tools: {names}"} | |
| tool_results = [] | |
| for block in tool_uses: | |
| yield {"type": "log", "text": f" -> {block.name}({json.dumps(block.input)})"} | |
| session = tool_sessions.get(block.name) | |
| if session is None: | |
| result_text = json.dumps({"error": f"Unknown tool: {block.name}"}) | |
| else: | |
| result = await session.call_tool(block.name, block.input) | |
| result_text = result.content[0].text if result.content else "" | |
| tool_results.append({ | |
| "type": "tool_result", | |
| "tool_use_id": block.id, | |
| "content": result_text, | |
| }) | |
| messages.append({"role": "assistant", "content": response.content}) | |
| if response.stop_reason == "tool_use": | |
| messages.append({"role": "user", "content": tool_results}) | |
| continue | |
| # Final answer | |
| answer = "\n".join( | |
| block.text for block in response.content if hasattr(block, "text") | |
| ) | |
| yield {"type": "result", "text": answer} | |
| break | |
| # --------------------------------------------------------------------------- | |
| # CLI shim — wraps the generator, prints to stdout | |
| # --------------------------------------------------------------------------- | |
| async def run_agent(question: str) -> None: | |
| api_key = os.environ.get("ANTHROPIC_API_KEY", "") | |
| async for event in run_agent_stream(question, api_key): | |
| if event["type"] == "log": | |
| print(event["text"]) | |
| elif event["type"] == "result": | |
| print("\n" + "=" * 70) | |
| print(event["text"]) | |
| print("=" * 70 + "\n") | |
| # --------------------------------------------------------------------------- | |
| # Entry point | |
| # --------------------------------------------------------------------------- | |
| def main() -> None: | |
| if len(sys.argv) > 1: | |
| question = " ".join(sys.argv[1:]) | |
| else: | |
| print("Market Intelligence Agent") | |
| print("-" * 40) | |
| question = input("Question: ").strip() | |
| if not question: | |
| print("No question provided.") | |
| sys.exit(1) | |
| asyncio.run(run_agent(question)) | |
| if __name__ == "__main__": | |
| main() | |