Review v1 · round 1 · manuscript v1
mcp-proto-okn: Natural-language access to open scientific knowledge graphs through the Model Context Protocol
Panel readout
4 of 5 specialists scored · 1–5
Legacy scaled score
Every referee landed on the same score.
- reporting reproducibility4.0Confidence 4 of 5
- data analysis4.0Confidence 3 of 5
- scientific validity4.0Confidence 4 of 5
- contribution context4.0Confidence 4 of 5
- ethicsThe reviewer gave no score and did not say why.n/a
Score is the referee's assessment of the work.Confidence is how sure that referee was of its own reading, recorded separately and never combined. The editor's verdict is its own judgment of the reports, not a threshold applied to this mean. The legacy aggregate is the historical panel mean multiplied by 20. It is not comparable to the Editor-in-Chief's publication readiness score. n/a means that referee found nothing in its dimension to judge here. It is left out of the mean rather than counted as a good score, which is what a forced number quietly became.
Abstract
as posted by the authors
MCP Server Proto-OKN (mcp-proto-okn) is a Python-based Model Context Protocol server that enables AI assistants to discover, inspect, query and integrate scientific knowledge graphs through natural language. The server provides graph routing, schema inspection, SPARQL execution, ontology expansion, multi-graph querying, and transcript generation, lowering the barrier to cross-domain knowledge graph analysis for biomedical and scientific users. mcp-proto-okn is implemented in Python using the FastMCP framework and is available at https://github.com/sbl-sdsc/mcp-proto-okn. Documentation, client configuration instructions, and example analysis transcripts are provided in the GitHub repository.
The review
- SummaryThe panel's assessment in brief.
- Decision letterThe editor's verdict and what it requires.
- Desk screenWhether the submission cleared the bar for full review.
- Advocate / skeptic debateThe case for and against, in full.
- Debate synthesisThe condensed account of the debate the editor read.
- Venue suggestionsWhere this might be submitted.
- Manuscript statisticsDeterministic counts over the text the panel read.
Specialist reports
Editorial audits
Factual checklists, not opinions. They skip the debate and go straight to the editor.
The text the panel read
counted, not judged
Counted at ingest, no model involved. These describe theconverted text the referees read, not your PDF. This conversion was flagged degraded, so the figures below are less reliable than usual.
Size
- Words
- 2,170
- Main text
- 1,987
- Sentences
- 80
- Display equations
- 0
excluding references
Sentences
- Median sentence
- 24.5 words
- Longest tenth
- 43 words
- Over 40 words
- 18%
- Passive
- ~0.2/sentence
regex approximation
Evidence on the page
- Citations
- not countable
- Numbers
- 41.47
- p-values
- 0 exact, 0 threshold
this venue most likely sets them as superscript numerals, which convert to bare digits
per 1000 words
Hedging against amplifying
Per 1000 words. Softening ("may", "suggests") against strengthening ("clearly", "demonstrates"). No referee saw these.
Provenance
PeerReviewAgents 0.5.0 · 1cf57690
Reviewed file
v1 · 421 KiB
sha256 17730e4ffd5ace71eb3e0921…
Debate rounds
2
Run cost
$0.84
Manuscript read as
markdown
Converted by rustypaper 0.2.0. Quotations match the manuscript text.
Conversion degraded: 84% of the text matched no known section — section-keyed statistics are unavailable. The panel read damaged text, and the damage is the converter's, not the authors'.
Desk screen
triage gate
Which model wrote which report
| Stage | Model |
|---|---|
| Editorial audits (×2) | claude-haiku-4-5 |
| Advocate / skeptic | claude-sonnet-5 |
| Specialist reviewers (×5) + desk screen | claude-haiku-4-5 |
| Editor, debate synthesizer | claude-opus-5 |
| debate synthesizer (override) | claude-sonnet-5 |
| journal recommender (override) | claude-haiku-4-5 |
The widest fan-out runs on the cheapest model. Only the agents that decide the verdict run on the most capable one.
What the referees looked up (16 searches)
| Agent | Search | Answered by | Hits |
|---|---|---|---|
| audit citation integrity | find related workEmonet LLM SPARQL federated knowledge graphs 2410.06062 | N/A | 3 |
| find related workKinjo TogoMCP MCP SPARQL RDF schema 2026 | N/A | 3 | |
| find related workMorris SPOKE scalable precision medicine knowledge graph 2023 | N/A | 3 | |
| find related workGao BioBricks versioned data registry life sciences 2408.17320 | N/A | 3 | |
| reviewer contribution context | find related workModel Context Protocol MCP knowledge graphs SPARQL | N/A | 5 |
| find related worknatural language SPARQL query generation LLM | N/A | 5 | |
| find related workProto-OKN knowledge graph federation | N/A | 5 | |
| search preprintsMCP Model Context Protocol scientific knowledge | N/A | 5 | |
| search preprintsTogoMCP natural language knowledge graph query | N/A | 1 | |
| find related workontology expansion MONDO UBERON hierarchical reasoning | N/A | 5 | |
| find related workfederated SPARQL query execution multiple endpoints | N/A | 5 | |
| search preprintsLLM SPARQL query generation federated knowledge graphs | N/A | 2 | |
| search preprintsEmonet SPARQL federated knowledge graphs RAG | N/A | 0 | |
| find related workSPOKE knowledge graph biomedical | N/A | 5 | |
| search preprintsOpen Knowledge Network OKN Fabric NSF | N/A | 0 | |
| find related worktext-to-SPARQL generation schema inspection | N/A | 5 |
Run against arXiv, Semantic Scholar, PubMed and bioRxiv while the review was being written. A search returning zero hits is kept: it is the evidence behind a referee saying it found no prior art.
What each agent cost
| Agent | USD |
|---|---|
| editor | $0.3044 |
| skeptic | $0.1314 |
| advocate | $0.1309 |
| debate synthesizer | $0.0813 |
| reviewer contribution context | $0.0600 |
| audit citation integrity | $0.0463 |
| audit methods completeness | $0.0237 |
| desk screen | $0.0152 |
| journal recommender | $0.0109 |
| reviewer data analysis | $0.0108 |
| reviewer scientific validity | $0.0099 |
| reviewer reporting reproducibility | $0.0090 |
| reviewer ethics | $0.0035 |
Cite this review
Permanent: this review only
This URL is a permanent link to this specific review, and will not change.
In Silico (2026). Review of "mcp-proto-okn: Natural-language access to open scientific knowledge graphs through the Model Context Protocol". In Silico. https://pgarrett-scripps.github.io/insilico/reviews/2026/mcp-proto-okn-natural-language-access-to-open-2605-30283/v1/
@misc{insilico-mcp-proto-okn-natural-language-access-to-open-2605-30283-v1,
title = {Review of {mcp-proto-okn: Natural-language access to open scientific knowledge graphs through the Model Context Protocol}},
author = {{In Silico}},
year = {2026},
howpublished = {In Silico, an AI-refereed overlay journal},
url = {https://pgarrett-scripps.github.io/insilico/reviews/2026/mcp-proto-okn-natural-language-access-to-open-2605-30283/v1/},
note = {Machine-generated peer review of doi:10.48550/arXiv.2605.30283 v1. Produced by PeerReviewAgents 0.5.0. Produced by PeerReviewAgents, doi:10.5281/zenodo.21781895.}
}Please cite the preprint itself as well. This reviews that work, it does not replace it. The review is machine-generated and advisory. If you are citing it as evidence about the paper, say so explicitly.