Frequently asked questions

The free tier doesn't require an API key — just install the MCP server with the snippet from the homepage. For Standard or Premium tier, sign in to your account dashboard at nullary.ai/dashboard after subscribing, and generate API keys from the Settings page.

All 7 modalities, all 31 MCP tools, full data access. Free tier is rate-limited to 30 queries/min and 1,000 queries/day with a maximum of 50 results per query. No API key required, no credit card required.

Yes. Upgrades take effect immediately. Downgrades take effect at the end of your current billing period. Annual plans pro-rate on cancellation.

Yes — academic users get 50% off Standard tier. Email team@nullary.ai with your .edu address for the discount code. Premium tier discounts available case-by-case for academic labs and non-profits.

Email team@nullary.ai within 30 days of your first payment. We refund the full amount, no questions asked. Applies to Standard and Premium tiers. Includes credit for any queries you've already run.

Credit card via Stripe. Enterprise customers can pay via ACH transfer or wire on annual invoices.

Any MCP-compatible client. Tested and documented for Claude Desktop, Cursor, and Claude Code. The protocol is standard, so any future MCP client will work without changes on our end.

Copy the JSON snippet from the homepage. Paste it into ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows). Restart Claude Desktop. The Nullary tools appear in your tool list. Full details in Docs.

Cursor's MCP support uses similar JSON configuration. See Docs for the exact path and snippet.

Run claude mcp add nullary https://mcp.nullary.ai/mcp from your terminal. Or add manually to your Claude Code config. See Docs.

Yes. The REST API at api.nullary.ai/v1 exposes the same data and tools. Useful for scripts, data pipelines, and non-MCP applications. See Docs.

No. The free tier is unauthenticated. Set the NULLARY_API_KEY environment variable to free-tier-no-key-required (or omit it entirely on some clients) and you're done.

Public databases: ChEMBL, PubChem BioAssay, BindingDB, DepMap, BioGRID ORCS, ClinicalTrials.gov, EudraCT, Drugs@FDA, Thera-SAbDab, FLAb, THPdb, PROTAC-DB, Retraction Watch, and others. See Coverage for the full source list and per-modality breakdown.

Refresh cadences vary by source: ClinicalTrials.gov daily, PubChem monthly, ChEMBL quarterly. Per-source last-refresh timestamps are shown on Coverage and in API response metadata.

The asymmetry reflects the public-database landscape, not a product limitation. Small-molecule drug discovery has 70+ years of accumulated public data. Oligonucleotide therapeutics are newer, and most failure data is proprietary or in patents. Enterprise tier supplementary materials extraction will substantially expand thin-coverage modalities. See Methodology for details.

Records from structured public databases come with extraction_confidence = 1.0 (no LLM interpretation). Enterprise tier records use LLM extraction followed by adversarial verification with Claude Opus and human review for thin-coverage modalities. See Methodology.

A score from 0 to 1 indicating how reliably the record was extracted. 1.0 = deterministic from structured source (ChEMBL, PubChem, etc.). 0.85-0.95 = LLM-extracted from semi-structured sources with verification. 0.70-0.85 = LLM-extracted from supplementary materials with verification. Records below 0.70 aren't surfaced.

Yes. Nullary monitors Retraction Watch and Crossref for new retractions. Affected records have source_retracted=true set with the retraction date and reason. Retracted records remain queryable but are flagged in API responses.

Yes — see Coverage for the full source list with licenses, refresh cadences, and per-modality contributions.

Yes for most sources. Most data is Creative Commons (CC-BY 4.0) or Public Domain, which allow commercial redistribution with attribution. A few sources require academic-use review for commercial use. Every API response includes source_license per record so you can filter or attribute appropriately. See Coverage for the full license matrix.

Yes for CC-BY-licensed sources (most of them). Cite the source database for each record. The DOI/PMID is included with every record for direct citation. Link back to nullary.ai is appreciated but not required.

CC-BY-SA 3.0 (used by ChEMBL) requires share-alike: derivative works using ChEMBL-sourced data must carry compatible licenses. If you're building a commercial database that aggregates Nullary results including ChEMBL-derived records, those records must remain CC-BY-SA in your database.

Yes for any reasonable use case. Caching for performance, building internal tools, training ML models — all fine. Re-distributing the entire dataset as a competing product isn't, even where licenses technically permit (most CC-BY sources). If unsure, email team@nullary.ai.

Yes. As long as you respect source licenses (most allow this with attribution). The license_per_record metadata makes compliance straightforward to automate.

31 MCP tools across the 7 modalities, plus cross-modality search and provenance tools. Full reference in Docs.

Use search_target_history(uniprot_id) to return all failed approaches against a target across all modalities. search_indication_history(indication) does the same for a disease indication. Both return small molecules, CRISPR, antibodies, peptides, PROTACs, oligonucleotides, and clinical trials in one unified response.

Yes. Most tools accept filters for: target (UniProt or gene symbol), compound class, measurement type, confidence threshold, indication, therapeutic area, trial phase, source database, and year range. Full parameter list per tool in Docs.

search_inactive_compounds returns compounds that failed to bind/inhibit their intended target. search_failed_selectivity returns compounds that did bind their intended target but also hit unwanted off-targets. Different failure modes, different queries.

Each record returns DOI and PMID of the source publication. Cite the source publication directly. Optionally also cite Nullary as the aggregation layer: "Negative results aggregated via Nullary (nullary.ai) on [date]."

Free: 30 queries/min, 1,000 queries/day, max 50 results per query.
Standard: 120 queries/min, 50,000 queries/day, max 500 results per query.
Premium: 600 queries/min, unlimited daily, max 5,000 results per query.

If you're on Free, you've exceeded 30 queries/min or 1,000 queries/day. Slow down or upgrade to Standard. The API returns a 429 status with retry-after headers. Free tier bursts to 2x briefly.

Free: 50. Standard: 500. Premium: 5,000. For larger result sets, paginate with the cursor parameter.

Pass cursor parameter from the previous response. The response includes next_cursor if more results exist. See Docs.

Typical: 50-200ms for queries returning < 100 results. Up to 800ms for complex cross-modality queries. P95 < 1.5s. See status.nullary.ai for live performance data.

Email (for account), Stripe customer ID, hashed API key, query telemetry (tool names, response sizes, timestamps), usage counts. We don't store query parameter contents, result selections, or non-essential personal data. See Security.

No. We log that a search_inactive_compounds call happened with a 50-result response, but we don't log which specific target or compound you searched for.

Primary infrastructure is EU (Germany / Frankfurt). Hetzner for compute, Neon for database (eu-central-1 region). US replication available for enterprise customers on request.

Yes. Customer email and billing data is the only personal data processed. Data deletion requests are honored within 30 days. See Security.

Yes. Email team@nullary.ai. Account and all associated data deleted within 30 days. Query telemetry retained anonymized for product analytics.

Vaccine modality is in active ingestion. Expected live in v1.1 (within 4-6 weeks of v1.0 launch).

Extraction of failure data from PDFs, supplementary materials, conference abstracts, and patents. Adds millions of records to modalities currently with thin public-DB coverage. Sequenced by customer demand signal from public-database usage. See Methodology.

Yes — patents are part of the Enterprise tier. Expired patents disclose millions of failed compounds that aren't aggregated anywhere else. High-value but slow to extract correctly.

Maybe. Email team@nullary.ai with the source you're interested in. We prioritize based on customer demand and license clarity.