Bypassing Document Ingestion: An MCP Approach to Financial Q&A

A team of researchers has published a paper proposing a novel approach to financial question-answering that bypasses traditional document ingestion. Instead of relying on retrieval-augmented generation (RAG), which involves chunking and searching documents, their method uses the Model Context Protocol (MCP) to give large language models (LLMs) direct, tool-like access to curated financial data APIs, such as those from LSEG (London Stock Exchange Group). This allows the AI to query live, structured data sources programmatically, treating them as tools it can call upon.

The researchers tested this MCP-based system on the FinDER benchmark, a standard for evaluating financial document understanding. Their approach excelled on the 'Financials' subset, achieving up to 80.4% accuracy on complex, multi-step numerical questions when the correct context was retrieved. This performance highlights a significant advantage for quantitative analysis, where precise, up-to-date figures from vendor systems are more reliable than information extracted from static documents.

However, the paper also provides crucial evidence on the limitations of this data-centric approach. It breaks down for questions requiring qualitative reasoning or specific context from lengthy documents, areas where traditional RAG still holds an edge. The work establishes a valuable baseline for MCP-based financial QA and offers a clear framework for when to use direct data access versus document retrieval, helping practitioners choose the right tool for the task.