BGPT (BioGPT)

BGPT is an evidence layer for scientific AI, designed to present scientific claims with supporting evidence, scope, and limitations instead of unsupported assertions. It operates as both a web resource and an MCP (Model Context Protocol) integration, making evidence-qualified scientific statements available to tools like Claude, Cursor, and other compatible AI assistants. Each claim is paired with one evidence-qualified sentence and linked to a full evidence page for deeper verification. The homepage organizes current evidence into six scientific domains: microbiome and infectious disease, genomics/evolution/development, neuroscience/behavior, human health/medicine, AI methods/biosecurity, and ecology/origins of life. Example claims span topics from jellyfish sleep regulation and urolithin A longevity to avian song complexity and LLM-assisted systematic reviews. Rather than merely presenting findings, BGPT scores papers for quality, novelty, and reproducibility, and explicitly states limitations next to each conclusion. This makes BGPT useful for researchers, clinicians, students, and anyone relying on AI-generated or rapidly summarized scientific information. Its emphasis on completeness and caution—acknowledging when evidence is mechanistic, underpowered, or model-derived—helps reduce over-citation and scientific misinformation. The service is particularly relevant for professionals working with AI assistants who need source-grounded, limitation-aware evidence.