Bio Research Lookup

Search scientific literature across PubMed, Semantic Scholar, bioRxiv/medRxiv, ClinicalTrials.gov, and OpenAlex.

Step 0 — Pre-flight: interpret and clarify the query

Before any API calls, parse the user's intent and reformulate it as a scientific search query.

Always show this block first:

🔍 Понял как: **[научная формулировка на русском]**
Термины для поиска: `term1 term2 term3`

Then decide:

When to ask a clarifying question (append after the interpretation block):

• Query contains multiple unrelated topics ("кофе, алкоголь и сон" → ask: "Это три отдельных поиска или вас интересует их совместное влияние?")
• Term is polysemous ("стресс" → ask: "Психологический стресс, физиологический (кортизол, HPA), или оба?")
• Scope unclear ("витамин D" → ask: "Вас интересует дефицит, дозировка, конкретное заболевание?")
• Mechanism vs. clinical evidence unclear → ask which is needed

When NOT to ask — just proceed:

• PMID / DOI / link
• Already contains MeSH or scientific terms
• Clear causal question ("интервальное голодание и инсулинорезистентность")
• User already specified study type ("найди мета-анализы про...")

Reformulation examples:

Step 1 — Translate query to English

All APIs are English-only. After Step 0 clarification, translate the final search terms.

Examples:

• "интервальное голодание" → "intermittent fasting"
• "микробиом кишечника" → "gut microbiome"
• "клинические испытания метформина" → "metformin clinical trials"

Use the translated English terms for all API calls below.

Step 2 — Classify and pick platform(s)

Platform Instructions

Semantic Scholar (broad search, psychology, health questions)

Search:

GET https://api.semanticscholar.org/graph/v1/paper/search?query={URL_ENCODED_QUERY}&fields=title,abstract,tldr,year,authors,citationCount,venue&limit=5

By paper ID (S2 ID or DOI):

GET https://api.semanticscholar.org/graph/v1/paper/{ID}?fields=title,abstract,tldr,year,authors,citationCount,externalIds

• Response includes tldr.text — a 1-sentence AI summary. Prefer tldr.text over abstract when available.
• citationCount indicates paper influence — highlight highly cited papers
• If response is 429 (rate limit) → fall back to OpenAlex (see below), then PubMed

OpenAlex (fallback for Semantic Scholar, 250M+ papers)

Use when Semantic Scholar returns 429 or empty results.

GET https://api.openalex.org/works?search={URL_ENCODED_QUERY}&per-page=5&select=title,abstract_inverted_index,publication_year,authorships,cited_by_count,primary_location&mailto=research@example.com

• abstract_inverted_index is an inverted index — reconstruct the abstract by sorting keys by their position values:

  { "word": [pos1, pos2], ... } → sort all (word, pos) pairs by pos → join words
  

• primary_location.source.display_name = journal name
• authorships[0].author.display_name = first author

PubMed (medical, clinical, NIH database)

Step 1 — Search for PMIDs:

GET https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&term={URL_ENCODED_QUERY}&retmax=5&retmode=json&tool=claude-skill&email=research@example.com

To prioritize high-quality evidence, append publication type filters:

• Meta-analyses: add +AND+Meta-Analysis[pt]
• RCTs: add +AND+Randomized+Controlled+Trial[pt]
• Reviews: add +AND+Review[pt]

Step 2 — Fetch abstracts by PMID(s):

GET https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&id={COMMA_SEPARATED_PMIDS}&rettype=abstract&retmode=text&tool=claude-skill&email=research@example.com

• If direct PMID provided, skip Step 1 and go straight to efetch
• Response is plain text. Look for PT lines (Publication Type) to extract study type for badges (see Step 4)

bioRxiv / medRxiv (preprints)

By DOI:

GET https://api.biorxiv.org/details/{server}/{DOI}/na

Where {server} = biorxiv or medrxiv

Recent papers by date range:

GET https://api.biorxiv.org/details/{server}/{YYYY-MM-DD}/{YYYY-MM-DD}

Filter by category: append ?category={topic} (e.g. neuroscience, cell_biology, epidemiology)

Response is JSON with abstract field in each item.

ClinicalTrials.gov (active and completed clinical trials)

Use when user asks about trials, ongoing studies, drug testing, or phase 1/2/3.

GET https://clinicaltrials.gov/api/v2/studies?query.term={URL_ENCODED_QUERY}&pageSize=5&format=json

Key fields from response (studies[].protocolSection):

• identificationModule.nctId → trial ID (e.g. NCT04727359)
• identificationModule.briefTitle → trial name
• statusModule.overallStatus → Recruiting / Completed / Active, not recruiting
• designModule.phases → ["PHASE2", "PHASE3"]
• descriptionModule.briefSummary → short description
• statusModule.startDateStruct.date → start date

Trial link: https://clinicaltrials.gov/study/{NCTId}

Step 3 — Fetch and parse results

Extract from whichever platform(s) used:

• Title
• Authors (first author + et al.)
• Year / date
• TLDR or abstract summary
• Citation count (if available)
• Publication type (for badge)
• Link to paper / trial

Step 4 — Present results

Study type badge (from PubMed PT field or inferred):

Paper format:

## [Badge] [Title] (Year)
**Авторы:** First Author et al.
**Журнал:** Journal/venue
**Цитирований:** N

[tldr.text if available — otherwise 2-3 sentence summary of key findings]

🔗 Link

ClinicalTrials format:

## 🏥 [Trial Title]
**Статус:** Recruiting / Completed / Active
**Фаза:** Phase 2 / Phase 3
**Начало:** Date

[Brief summary 1-2 sentences]

🔗 https://clinicaltrials.gov/study/{NCTId}

If multiple results, list top 3–5 ranked by relevance and citation count.

After listing papers — add synthesis block:

---
## Что говорит наука

[2-4 sentences synthesizing the overall evidence: is there consensus or conflict,
what is the effect size / str