Author: Axiomatic AI Date: September 8, 2025
<aside>
We present a case study showing how researchers in quantum chemistry and quantum simulation can reproduce and extend literature results using an agentic workflow inside Cursor, connected to domain tools via Model Context Protocols (MCPs). The workflow automates three recurring tasks: (i) extracting methods and parameters from a reference paper, (ii) orchestrating computationally demanding electronic-structure and quantum-computing simulations across libraries, and (iii) comparing against baseline results. The outcome is a reusable agentic workflow that reduces setup friction, automates orchestration of computational tasks and results analysis, and accelerates benchmarking across multiple algorithmic choices and configurations.
</aside>
Researchers in computational sciences, such as quantum chemistry and quantum simulation, frequently need to reproduce results from the literature and benchmark alternative approaches (e.g., different VQE ansätze, circuit depths, and basis sets) on the same molecular systems. Doing this manually typically requires writing and maintaining glue scripts across multiple libraries, along with significant effort for data wrangling and plotting. These tasks can be particularly daunting for beginner researchers, who must juggle complex scripting, data handling, and visualization before they can even focus on the science.
AI agents offer an opportunity to automate many of these tasks, helping experienced researchers accelerate their work and enabling newcomers to enter the field more easily. To achieve this, however, such agents must be paired with domain-specific tools. Model Context Protocols (MCPs) provide a standard for connecting these tools into agentic frameworks, such as AI-powered IDEs.
In this case study, we leveraged Cursor — an AI-powered IDE —connected to MCPs for document analysis (developed by Axiomatic AI), as well as open-source MCPs for PySCF [1], a quantum chemistry package, and Qiskit [2], a quantum computing framework. We used these tools to first reproduce the results from a paper implementing the Variational Quantum Eigensolver (VQE) to compute the electronic energy of a molecular system (H₂), and then extended the workflow to explore alternative ansätze and compare the outcomes against those reported in the paper. In total, we set up, ran, and analyzed over 300 distinct energy calculations by guiding an AI agent with natural-language prompts, without manually writing any code.
We used the chat agent in Cursor as an orchestrator agent. This agent executed a plan that coordinated different tools via MCPs to achieve the overall goal. We selected GPT-5 as the model for the agent.
The following MCPs were used to parse information from papers, perform quantum chemistry calculations, and run quantum algorithms:
Provides a tool to parse PDF content into Markdown, facilitating extraction of data from tables, and extracting figures.
AxDocumentParser.document_to_markdown
Extracts relevant information, equations, parameters, and contextual descriptions to provide comprehensive analysis of document content. We used it to extract details from figures (e.g., a quantum circuit) and then use the description to replicate the circuit using the tools in the qiskit-mcp-server.
AxDocumentAnnotator.annotate_pdf