Dan Gause

Post-baccalaureate researcher at NRAO studying protostellar outflows and building software tools for astronomy.


Session

11-03
10:36
12min
MANNA: an MCP Architecture for NOIRLab, NRAO, and additional Archives
Dan Gause

Astronomical data archives are an invaluable resource for research in modern astronomy. While many archives implement the same set of standards defined by the International Virtual Observatory Alliance (IVOA), differences in underlying data and architectural implementations necessitate significant archive-specific knowledge. This requirement can impede research, especially in multiwavelength astronomy. AI agents offer a promising way to lower this barrier, but without structured access to archive services and metadata they must rely on general knowledge or web searches, which are often incomplete or outdated for a given archive, and the searching itself costs time and tokens. To address this problem, we developed MANNA (MCP Architecture for NOIRLab, NRAO, and additional Archives), a Model Context Protocol (MCP) server tailored to IVOA-based astronomical archives that provides tools for archive discovery and data querying. We found that MANNA improved performance across an evaluation suite of one open-weight and three Anthropic Claude models, increasing efficiency for all models and substantially improving accuracy for the smallest Claude model. This effect was magnified when limits on turns and time per task were applied to evaluated tasks, supporting the practical use of MANNA under budgetary or time-related pressure. Although MANNA was developed mainly for the NOIRLab and NRAO archives, it is a flexible framework that is extensible to any IVOA-based archive.

AI as a tool for data discovery and data management
Seminar Room 2