New benchmark with 613 financial tasks tests how well AI language models can use real financial tools to solve practical problems

FinMCP-Bench contains 613 samples across 10 main scenarios and 33 sub-scenarios with both real and synthetic financial queries

Benchmark integrates 65 real financial MCPs (Model Context Protocols) and evaluates three complexity levels: single-tool, multi-tool, and multi-turn tasks

Introduces specialized metrics to measure tool invocation accuracy and reasoning capabilities of mainstream LLM agents in financial applications

Provides standardized testbed for advancing research on financial AI agents solving real-world problems