LLM semantic understanding > regex keyword matching. Data uses read_draft MCP tool; review logic is handled by LLM + domain knowledge.
Trigger: review draft, reviewer, CONSORT, STROBE, PRISMA, CARE, pre-submission check
Paired with MCP Tools
| Tool | Purpose |
|---|---|
read_draft |
Required — read draft |
list_drafts |
List drafts available for review |
count_words |
Word count |
check_formatting |
Journal format quantitative check + cross-document consistency |
Framework 1: Peer Review
Use read_draft to retrieve content → review across 4 dimensions → output structured report
Review dimensions
1. Structure: required sections complete, structured Abstract, logical order (Major if missing/out of order)
2. Methods:
| Item | Major/Minor |
|---|---|
| Ethics/IRB statement | Major |
| Statistical methods | Major |
| Sample size justification | Minor |
| Primary outcome definition | Minor |
| Inclusion/exclusion criteria | Minor |
3. Statistics:
| Item | Major/Minor |
|---|---|
| Confidence intervals | Major |
| P-value reporting (exact values) | Minor |
| Multiple comparisons correction | Minor |
| Effect size | Minor |
4. Writing:
| Item | Major/Minor |
|---|---|
| Limitations section | Major |
| Causal language in observational | Major |
| Word count / Reference density | Minor |
Judgment
Major ≥3 → 🔴 Major revision required | 1-2 → 🟡 Major revision | 0 → ✅ Minor/Accept
Framework 2: Reporting Guideline Compliance
read_draft → automatically detect/specify guideline → item-by-item check → compliance report
Auto-Detection
| Draft characteristic | Guideline |
|---|---|
| randomized, RCT, blinded, placebo | CONSORT |
| systematic review, meta-analysis | PRISMA |
| case report, case presentation | CARE |
| observational, cohort, case-control | STROBE |
CONSORT (RCT, 21 items)
Title(1), Abstract(2), Intro(3-4: background, objectives), Methods(5-13: design, eligibility, interventions, outcomes, sample size, randomization, allocation concealment, blinding, statistics), Results(14-19: flow, recruitment, baseline, N analysed, outcomes+CI, harms), Discussion(20: interpretation+limitations), Other(21: registration)
STROBE (Observational, 16 items)
Title(1), Intro(2-3), Methods(4-11: design, setting, participants, variables, measurement, bias, sample size, statistics), Results(12-14: participants, descriptive, outcome), Discussion(15: interpretation), Other(16: funding)
PRISMA (SR, 17 items)
Title(1), Abstract(2), Intro(3), Methods(4-11: protocol, eligibility, sources, strategy, selection, extraction, RoB, synthesis), Results(12-15: flow, characteristics, results, heterogeneity), Discussion(16), Other(17: registration)
CARE (Case Report, 13 items)
Title(1), Abstract(2), Intro(3), Patient(4-7: demographics, complaints, history, exam), Clinical(8-10: diagnosis, intervention, follow-up), Discussion(11-12: strengths/limits, rationale), Patient(13: consent)
Judgment
≥80% → 🟢 Good | 60-80% → 🟡 Needs attention | <60% → 🔴 Significant gaps
Usage principles
- Semantic understanding, not keyword matching ("ethics committee approval" = ethics approval)
- Major = may be rejected, Minor = suggested improvement
- Give concrete fix suggestions
- Multi-file support (Introduction and Methods may be in different files)