Developer Platform

GaiaLab as an MCP Server

Nine biomedical intelligence tools, callable from Claude Desktop, Cursor, ChatGPT, or any MCP-compatible agent. No auth for standard analyses. Evidence-verified outputs.

POST https://www.gailabai.com/mcp
9
MCP Tools
76+
Data Sources
0%
Auth for Standard Analyses
PMID
Claim Verification
Quick Start

Connect in 60 seconds

Add GaiaLab to any MCP client. No API key, no sign-up. Discovery doc at /.well-known/mcp.json.

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "gaialab": {
      "url": "https://www.gailabai.com/mcp",
      "transport": "streamable-http"
    }
  }
}

Cursor / VS Code (.cursor/mcp.json)

{
  "servers": {
    "gaialab": {
      "url": "https://www.gailabai.com/mcp"
    }
  }
}

Raw HTTP (any language)

curl -X POST https://www.gailabai.com/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "method": "tools/call",
    "params": {
      "name": "gaialab_get_target_brief",
      "arguments": {
        "gene": "KRAS",
        "diseaseContext": "pancreatic cancer"
      }
    }
  }'

Discovery document

curl https://www.gailabai.com/.well-known/mcp.json

Returns tool list, endpoint, and capability declaration in one JSON object. MCP clients use this for auto-configuration.

Tool Reference

9 Biomedical Intelligence Tools

All tools return JSON. Evidence-verified outputs include grounded ratio, cite-F1, and PMID hallucination rate where applicable.

gaialab_generate_insights

Full gene-panel analysis: enriched pathways (FDR-corrected), therapeutic strategies, drug candidates (Tier I/II/III), novel hypotheses, and a 16-channel evidence brief. The primary analysis tool.
genesrequiredArray of HGNC gene symbols, e.g. ["BRCA1","TP53","PIK3CA"]
diseaseContextrequiredDisease or biological context, e.g. "triple-negative breast cancer"
audienceoptionalresearcher | clinician | executive | student (default: researcher)
includeDrugsoptionalBoolean, default true. Set false for faster pathway-only analysis.

gaialab_compare_targets

Runs analysis on two gene panels in parallel and returns a structured diff: shared pathways, unique pathways, drug overlap, grounded ratio delta. Ideal for comparing disease subtypes, mutation profiles, or treatment contexts.
panelA.genes + panelA.diseaseContextrequiredFirst panel gene list and disease context
panelB.genes + panelB.diseaseContextrequiredSecond panel gene list and disease context

gaialab_verify_claim

Evidence integrity check for any biomedical claim. Runs PMID existence verification against PubMed E-utilities and zero-shot NLI entailment scoring (active model published live at /api/nli/status). Returns cite-F1, hallucination flags, and evidence status.
claimrequiredThe biomedical statement to verify
pmidsoptionalArray of PMIDs to validate against PubMed
geneContextoptionalGene symbols for biological grounding

gaialab_diff_snapshots

Compares two saved GaiaLab analyses: new/removed pathways, drug candidate shifts, hypothesis confidence changes, quality score deltas. Use gaialab_search_snapshots to find IDs.
snapshotIdArequiredID of the earlier snapshot
snapshotIdBrequiredID of the later snapshot

gaialab_get_target_brief

Structured one-page target intelligence brief for a single gene: protein function, disease associations, AlphaFold structural druggability score, top drug candidates with ADMET, active clinical trials, and top FDR-significant pathways. Top drug candidates require a Researcher-tier API key or higher — everything else works unauthenticated; an unauthenticated call returns topDrugs: [] with topDrugsNote explaining why (see /pricing).
generequiredHGNC gene symbol, e.g. KRAS, EGFR, CDK4
diseaseContextoptionalDisease focus (default: cancer)

gaialab_prioritize_repurposing

Returns ranked drug repurposing candidates for a gene panel and disease. Each candidate includes tier (I/II/III), 6-factor score breakdown, ADMET profile (Lipinski, QED, MW, LogP), and evidence quality flag. Requires a Researcher-tier API key or higher — an unauthenticated/free-tier call returns an empty candidate list with note explaining why (see /pricing).
genesrequiredGene panel array
diseaseContextrequiredDisease context
tierFilteroptionalI | II | all (default: all)

gaialab_generate_validation_plan

Generates a step-by-step experimental validation roadmap for a scientific hypothesis: in vitro assays → in vivo models → translational clinical studies. Each stage includes assay type, expected outcome, critical controls, timeline, and go/no-go criteria.
hypothesisrequiredThe hypothesis to validate, in plain scientific language
geneContextoptionalRelevant gene symbols
diseaseContextoptionalDisease context

gaialab_search_snapshots

Searches the GaiaLab analysis archive by disease context, gene symbol, or recency. Returns snapshot IDs, timestamps, gene panels, and quality scores. Use returned IDs with gaialab_diff_snapshots.
diseaseoptionalPartial match on disease context, e.g. "breast cancer"
geneoptionalGene symbol present in the panel, e.g. "TP53"
limitoptionalMax results (1–50, default 10)

gaialab_deep_research

One call that chains internally: (1) full gene-panel analysis across 76+ databases, (2) NLI + PMID hallucination verification of the top insights, and (3) an AI-generated experimental validation roadmap for the top hypothesis. Returns a unified JSON: analysis summary, verified claims with entailment scores, and a step-by-step in-vitro → in-vivo → clinical validation plan. Drug candidate data requires a Researcher-tier API key or higher — everything else works unauthenticated; an unauthenticated call returns empty drugCandidates with drugAnalysisNote explaining why (see /pricing).
genesrequiredArray of HGNC gene symbols (1–50)
diseaseContextrequiredDisease or therapeutic area, e.g. "triple-negative breast cancer"
researchQuestionoptionalA specific question to guide the analysis and validation plan
audienceoptionalresearcher | clinician | executive (default: researcher)
Example Workflow

Target discovery in 3 tool calls

A typical translational researcher workflow using GaiaLab as a reasoning backend.

Step 1 — Get a target brief
// Tool: gaialab_get_target_brief
{ "gene": "PARP1", "diseaseContext": "BRCA1-mutant ovarian cancer" }

// Returns (real output): proteinName="Poly [ADP-ribose] polymerase 1",
// molecularWeight=113084 Da, alphaFoldPlddt=82.38 (high confidence),
// druggabilityScore=0.6 (0-1 scale; domain + enzyme-activity based structural score),
// topDrugs=[] with topDrugsNote explaining a Researcher-tier key is required for
// ranked drug candidates on unauthenticated/free-tier calls (see /pricing),
// top pathways: base excision repair (q=0.0033), apoptosis (q=0.0033),
// NF-κB signaling (q=0.0033)
Step 2 — Verify a hypothesis
// Tool: gaialab_verify_claim
{
  "claim": "PARP1 inhibition causes synthetic lethality in BRCA1-deficient cells",
  "pmids": ["15829967", "15829966"],
  "geneContext": ["PARP1", "BRCA1"]
}
// pmids are Farmer et al. 2005 and Bryant et al. 2005 (Nature) -- the original
// papers establishing PARP-inhibitor synthetic lethality in BRCA-deficient cells.

// Returns (real output): pmidsChecked=true, hallucinated=0, evidenceStatus="supported",
// citeF1=1, nliEnabled=true, summary="All 2 cited PMIDs verified as real in PubMed.
// Textual support for the claim: supported."
Step 3 — Generate validation plan
// Tool: gaialab_generate_validation_plan
{
  "hypothesis": "PARP1 inhibition causes synthetic lethality in BRCA1-deficient cells",
  "geneContext": ["PARP1", "BRCA1"],
  "diseaseContext": "BRCA1-mutant ovarian cancer"
}

// Returns: 4-stage plan: cell-line viability assay → clonogenic survival →
// xenograft model → Phase II basket trial design, each with
// controls, timeline, and go/no-go criteria
All three calls take <60 seconds combined. Results include snapshot IDs for reproducibility. Diff two runs with gaialab_diff_snapshots to track how evidence evolves over time.
Data Sources

76+ live databases, synthesized in parallel

Every tool call fetches live data. No stale knowledge cutoffs.

Literature
PubMed, PMC Full-Text, Semantic Scholar, bioRxiv
Pathways
KEGG, Reactome, Enrichr, Gene Ontology
Drugs
ChEMBL, DGIdb, PharmGKB, DrugCentral, FDA
Disease
OpenTargets, DisGeNET, Monarch / OMIM / HPO
Structure
AlphaFold EBI, UniProt, STRING, BioGRID
Clinical
ClinicalTrials.gov v2, cBioPortal, CIViC, OncoKB
API Keys

Get your API key

Free tier: 10 analyses/day. Keys are optional — unauthenticated requests share a rate-limited pool.

Free self-serve key generation is not available. Request a Researcher key via the academic program or a paid plan on the pricing page.

Check usage
# Check quota and usage stats
curl -H "X-API-Key: gl_..." https://www.gailabai.com/api/keys/usage
Embeddable widget
<!-- Drop-in embed -->
<div id="gaialab-widget"
     data-genes="BRCA1,TP53,PTEN"
     data-disease="triple-negative breast cancer"></div>
<script src="https://www.gailabai.com/widget.js" async></script>

// Programmatic:
GaiaLab.render({
  target: '#my-container',
  genes:  ['KRAS', 'STK11'],
  disease: 'KRAS-mutant NSCLC',
  apiKey:  'gl_...',
  onReady: (info) => console.log('Snapshot:', info.snapshotId)
});
Evidence Integrity

How GaiaLab verifies its own outputs

Every analysis runs a 3-stage evidence integrity pipeline before returning results.

Stage 0 — PMID hallucination check

All AI-generated PMIDs batch-queried against NCBI E-utilities esummary. Hallucinated citations stripped before display. Rate reported on every result.

Stage 1 — NLI entailment (facebook/bart-large-mnli)

Claims verified against abstract text using a zero-shot NLI model. Entailment threshold 0.45. 1,500-char passage window. Live status: /api/nli/status.

Stage 2 — ALCE cite-F1

Citation precision, recall, and F1 computed per ALCE benchmark methodology. cite-F1 ≥ 0.6 = well-grounded. Reported on every trust snapshot.

See full methodology · live trust dashboard · benchmark scores