DBCLS BioHackathon 2026 report: Extending TogoMCP beyond RDF Portal while making its schema guides check their own answers
Abstract
TogoMCP is a Model Context Protocol (MCP) server that lets large language model (LLM) agents query life-science knowledge graphs in SPARQL, guided by per-database schema documents called MIE (Metadata Interoperability Exchange) files. During the DBCLS BioHackathon 2026 (BH26), the TogoMCP group set out to extend the server, refine the MIE files, add databases, learn what makes a good SPARQL example, and turn use cases into reusable agent skills. In the first six days of the event we published eleven releases (v2.12.2 to v2.20.0) and grew the catalogue from 37 to 45 databases: Fanta.bio, WikiPathways, IDSM, PubCaseFinder, LIPID MAPS, SwissLipids and MarpolBase, plus BH26 Microbes, an experimental dataset of KEGG Orthology assignments for 57.6 million prokaryotic proteins built for the BioHackathon. Five of them are served from endpoints outside RDF Portal, and a sixth from a QLever engine inside it rather than Virtuoso, and they broke assumptions the rest of the corpus had taught, such as named-graph pinning, federation with `SERVICE`, literal typing, and even that a result set is complete. For the first time, MIE files were written by people other than the server's maintainer, one of them by the maintainer of the database it describes, and a second group member built TogoCX, a companion MCP server that returns database edges with instructions on how their claims may be stated. We made every worked example in the MIE corpus (now 420) assert its recorded result against the live endpoint in continuous integration. The first full comparison found drift that execution-only checks had missed, most seriously a NANDO release that made an example join silently miss about 88% of mapped diseases. Onboarding the new databases uncovered a series of quantified "silent wrong answers", queries that return plausible results rather than errors, including a category count inflated 3.1-fold by an upstream data defect and a gene enumeration that returns 10,000 of 18,080 rows without an error or a warning. We also added two PubCaseFinder tools for phenotype-driven rare-disease diagnosis support, began serving analysis workflows (agent skills) from the server, fixed failure modes found in production call logs, made TogoID errors suggest working conversion routes, and prepared a LOTUS natural-products graph for hosting on RDF Portal. We argue that for LLM-facing schema documentation, "the example still runs" is not evidence that it is still correct.
Related
- DBCLS BioHackathon 2026 report: Variant representation in RDF for precision medicine
- DBCLS BioHackathon 2026 report: QPX: Pathway analysis environment for non-model organisms
- DBCLS BioHackathon 2026 report: Mass spectrometry data analysis workflow and visualization towards trans-omics research
- DBCLS BioHackathon 2026 report: An interactive, harmonised BioSample metadata resource for biological sample discovery
- DBCLS BioHackathon 2026 report: Open, Reproducible Extraction of Data-Entity References from the Biomedical Literature – with Entry-Level Resolution via Identifiers.org / TogoID
BibTeX
@misc{kinjo2026dbcls,
title = {{DBCLS BioHackathon 2026 report: Extending TogoMCP beyond RDF Portal while making its schema guides check their own answers}},
author = {Akira R Kinjo and Shuichi Kawashima and Takatomo Fujisawa and Yasunori Yamamoto},
howpublished = {BioHackrXiv},
year = {2026},
month = sep,
doi = {10.37044/osf.io/t25ng_v2},
url = {https://osf.io/preprints/biohackrxiv/t25ng_v2/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - DBCLS BioHackathon 2026 report: Extending TogoMCP beyond RDF Portal while making its schema guides check their own answers
AU - Akira R Kinjo
AU - Shuichi Kawashima
AU - Takatomo Fujisawa
AU - Yasunori Yamamoto
PY - 2026
DA - 2026/09/18
PB - BioHackrXiv
DO - 10.37044/osf.io/t25ng_v2
UR - https://osf.io/preprints/biohackrxiv/t25ng_v2/
AB - TogoMCP is a Model Context Protocol (MCP) server that lets large language model (LLM) agents query life-science knowledge graphs in SPARQL, guided by per-database schema documents called MIE (Metadata Interoperability Exchange) files. During the DBCLS BioHackathon 2026 (BH26), the TogoMCP group set out to extend the server, refine the MIE files, add databases, learn what makes a good SPARQL example, and turn use cases into reusable agent skills. In the first six days of the event we published eleven releases (v2.12.2 to v2.20.0) and grew the catalogue from 37 to 45 databases: Fanta.bio, WikiPathways, IDSM, PubCaseFinder, LIPID MAPS, SwissLipids and MarpolBase, plus BH26 Microbes, an experimental dataset of KEGG Orthology assignments for 57.6 million prokaryotic proteins built for the BioHackathon. Five of them are served from endpoints outside RDF Portal, and a sixth from a QLever engine inside it rather than Virtuoso, and they broke assumptions the rest of the corpus had taught, such as named-graph pinning, federation with `SERVICE`, literal typing, and even that a result set is complete. For the first time, MIE files were written by people other than the server's maintainer, one of them by the maintainer of the database it describes, and a second group member built TogoCX, a companion MCP server that returns database edges with instructions on how their claims may be stated. We made every worked example in the MIE corpus (now 420) assert its recorded result against the live endpoint in continuous integration. The first full comparison found drift that execution-only checks had missed, most seriously a NANDO release that made an example join silently miss about 88% of mapped diseases. Onboarding the new databases uncovered a series of quantified "silent wrong answers", queries that return plausible results rather than errors, including a category count inflated 3.1-fold by an upstream data defect and a gene enumeration that returns 10,000 of 18,080 rows without an error or a warning. We also added two PubCaseFinder tools for phenotype-driven rare-disease diagnosis support, began serving analysis workflows (agent skills) from the server, fixed failure modes found in production call logs, made TogoID errors suggest working conversion routes, and prepared a LOTUS natural-products graph for hosting on RDF Portal. We argue that for LLM-facing schema documentation, "the example still runs" is not evidence that it is still correct.
ER -