Skip to content
All library documents

Building Quantitative Research Agents with Tool Calling and MCP

Article vn.py community
Read the original

Summary

This event outline explains how to extend a language model from generating strategy text to carrying out quantitative research tasks. It introduces tool calling through the OpenAI library: describe available functions to the model, interpret its requested calls, execute those functions locally, return their results, and preserve messages across multiple turns. The stated aim is to reduce the manual steps involved in extracting generated code and running a backtest.

The outline then describes a VNAG-based tool system, with a common schema for tools and wrappers for local Python functions. A ReAct-style loop would let an agent perform successive actions, including data queries and backtest runs, while MCP provides a standardized route to external tools. This is an agenda for a planned community session, not a technical tutorial or evaluation: it includes no implementation details, trading strategy, or measured evidence that the proposed agent workflow works reliably.

Key ideas

  • Tool calling lets a model request local functions, receive their results, and continue across multiple turns.
  • A quantitative research agent can wrap data queries and backtest execution as callable tools.
  • The outline proposes a shared tool schema and a ReAct-style loop for successive tool use.
  • MCP is presented as a standard way to connect agents with external tools.
  • The document is an event outline and supplies no implementation results or reliability assessment.

Tags

Machine learningBacktesting

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.