Loading ToolBench
Biotech diligence workspace with a glass-box research agent (Sonny): live reasoning, balanced and fully cited reports, plus wet-lab workbenches for IHC, flow cytometry, and western blot.
LUMINA exposes 23 biotech research tools with inconsistent quality. Most tools have descriptions (good), but parameter schemas are incomplete or missing across the board. Many tools lack input parameter type definitions, descriptions, or both. Tool names follow verb_noun convention reasonably well (search_*, get_*, analyze_*), but parameter documentation is sparse. Output schemas are not documented. Error handling and recovery guidance are absent. The server appears to be a STDIO-only MCP implementation, which caps protocol readiness severely.
Analyze competitive landscape for a therapeutic area
Analyze efficacy and safety data from clinical trials
Analyze company financials (revenue, burn rate, runway, cash position)
Benchmark trial design against similar studies
Calculate company valuation using DCF, comparable companies, or precedent transactions
Estimate clinical trial success rates using Gosset.ai-style pharmaceutical intelligence (simulated data for demonstration)
Get comparable biotech companies and their valuations
Parameter schemas incomplete or missing across all tools. Many parameters lack type definitions (string, number, boolean, array, object). Tools like gtex_expression, human_protein_atlas, and gosset_estimate_success have parameters with no type info visible in source.
Output schemas not documented. No tool explicitly declares what fields it returns or their types. LLMs cannot plan downstream tool calls or extract structured data without knowing response shape.
No error handling or recovery guidance. Tools lack descriptions of failure modes, retryability, or actionable error messages. Agents cannot self-correct on failures.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Get recent industry news and press releases
Get market size data for a therapeutic area or technology
Get SEC filings (10-K, 10-Q, 8-K, S-1) for a company
Get stock price and performance data
Get comprehensive trial benchmarks including PTRs, sample sizes, and endpoints from pharmaceutical database (simulated Gosset.ai data)
Get detailed information about a specific clinical trial
Predict trial success probability using AI model
Get Phase Transition Rates (PTRs) for indication/modality
Search pharmaceutical database for drug trials with verified outcomes
GTEx v8 normal-tissue expression - the authoritative baseline for how much a target is expressed across ~54 healthy human tissues (median TPM). Use this to quantify normal-tissue / off-tumor expression that defines an ADC or targeted-therapy safety window, and to judge tumor-vs-normal selectivity against a normal baseline.
Human Protein Atlas IHC normal-tissue antibody-scored protein expression - the tissue-level ground truth for protein distribution across ~44 normal tissues. Use this to quantify protein-level selectivity and safety window, especially where RNA is misleading (post-translational regulation, protein stability, localization).
Search ClinicalTrials.gov for trials by indication, drug, company, or NCT ID
Search for company information, pipeline, and recent news
Search PubMed for research publications
Track M&A deals and partnerships in biotech/pharma
Search the user's uploaded documents (${names}) - internal reports, papers, filings, or data the user attached to this run. Returns passages as citable evidence (id DOC:...). Use it to ground claims in the user's own material alongside public sources.
Parameter descriptions are generic or missing context. Many params lack guidance on valid ranges, formats, or dependencies. E.g., 'Gene symbol' vs 'Gene symbol (e.g., TP53, BRCA1; case-insensitive; required for expression lookup)'.
No pagination or result-limiting guidance. Tools like search_pubmed, search_clinical_trials, and search_company_intel accept 'limit' but do not document max values, default behavior, or whether results are sorted/ranked.