# NovoMCP

> The open computational chemistry engine for drug discovery and materials science.

NovoMCP is an MCP + REST engine that any MCP-compatible AI assistant can call — Claude Desktop, Cursor, Codex, Zed, Cline, ChatGPT, Gemini, and others — plus any pipeline that speaks HTTP. 69 tools in the catalog, 14 work with a fresh clone and no wiring; the rest unlock as you wire optional compute services (docking, MD, QM, structure prediction, ADMET). One engine, two domains: drug discovery runs as an 11-stage autonomous funnel; materials science as a flexible four-workflow toolkit. Apache-2.0 (orchestration core BSL 1.1 → Apache 2029-07-12). The FAVES compliance API is a separate hosted service for regulated pipelines.

## Company

- Company site: https://novomcp.com
- Drug discovery: https://novomcp.com/drug-discovery
- Materials science: https://novomcp.com/materials-science
- GitHub: https://github.com/NovoMCP/novomcp
- Documentation: https://docs.novomcp.com/

## Product

### The Engine (open source)
- Product page: https://novomcp.com/engine
- Repo: https://github.com/NovoMCP/novomcp (Apache-2.0; orchestration core BSL 1.1 → Apache 2029-07-12)
- Docs: https://docs.novomcp.com/
- MCP + REST server, self-hostable, runs on your hardware
- Install: `git clone` + `pip install` + `python main_https.py` (Python 3.10+), or `docker compose up`
- 69 tools in the catalog; 14 always-available on a fresh clone — 6 RDKit cheminformatics, 2 platform, 4 public-API searches (PubMed literature, ChEMBL, ClinicalTrials.gov, bioRxiv), 2 autonomous discovery triggers (run_novo_ag + run_materials_ag); the rest unlock with optional compute services
- Pluggable — auth, credit metering, audit sinks, LLM providers (OpenAI / Anthropic / Ollama / Azure)
- Works with any MCP-compatible AI assistant (Claude, Cursor, Codex, Zed, Cline, ChatGPT, Gemini, Grok, Mistral, Cohere, Ollama, VS Code, Windsurf, Continue.dev, Cody, OpenCode, Open WebUI, NVIDIA NeMo Agent Toolkit)
- Lightweight version: novomcp-lite (Apache-2.0, `pip install novomcp-lite`, on PyPI) — just the RDKit cheminformatics tools + public-API search (9 zero-config tools, incl. PubMed literature) as a Python library or MCP server, for when you want the wrappers rather than the full engine. Same code the engine runs on, minus the platform. https://github.com/NovoMCP/novomcp-lite

### REST API (Live)
- Product page: https://novomcp.com/api · Docs: https://docs.novomcp.com/
- Base URL: `https://api.novomcp.com/v1` — every tool is `POST /v1/tools/{name}` with `{"arguments":{…}}` and a Bearer key
- Live OpenAPI 3.1 spec: https://api.novomcp.com/openapi.json (69 MCP tools + 1 API-only tool, schemas, x-tier/x-credits/x-compute/x-mcp-exposed)
- Auth: Bearer API key. Self-hosted engines mint their own keys via the auth service; cloud deployments get keys provisioned during onboarding.
- Bring-your-own-agent: expose the engine's tools to your own LLM (Claude API, GPT, anything) via HTTP
- Compute tools (docking, MD, QM, NNP, structure) require the corresponding services to be wired

### FAVES compliance API (hosted)
- Product page: https://novomcp.com/faves
- URL: `https://api.novomcp.com/v1/faves/check`
- Regulatory compliance API for pharmaceutical and biotech pipelines
- DEA scheduling, EU REACH, PAINS filters, structural alerts, sponsor whitelists, PMDA/KFDA/TGA
- SLA + immutable audit-log retention + drift monitoring + IQ/OQ/PQ documentation for 21 CFR Part 11
- Closed and paid — access on request. Not shipped with the OSS engine.

### NovoWorkbench (Desktop, design-partner preview)
- Product page: https://novomcp.com/novoworkbench
- Native molecular workstation for macOS and Windows
- 3D structure viewer, local RDKit compute, offline alert screening (PAINS + druglikeness rule sets), bring-your-own-LLM chat
- Currently in design-partner preview; ships as OSS in v1.5.x per the product roadmap
- Access: https://novomcp.com/sales-inquiry

### Novo Compute (GPU / QM / MD / FEP surface)
- Product page: https://novomcp.com/novo-compute
- GPU docking, MD simulation, QM, NNP, structure prediction, materials-science workflows
- Runs on your cluster (Kubernetes / Docker Compose reference deploys) or on ours (cloud deployment)
- Shares the same tool namespace as the core engine — same call pattern

### Extensions (sideload surfaces)
- Extensions page: https://novomcp.com/extensions
- Chrome extension: LIVE on the Chrome Web Store — https://chromewebstore.google.com/detail/novomcp/nfnhfmfmaccidmepfhllcllnjopeonmf
  - Molecular intelligence for PubChem, ChEMBL, bioRxiv, medRxiv, arXiv
  - Hover any SMILES for profile cards, ADMET, similarity search
  - Repo: https://github.com/NovoMCP/novomcp-chrome-sideload
- Word add-in: sideload via https://addin.novomcp.com/manifest.xml
  - Repo: https://github.com/NovoMCP/novomcp-word-sideload

### API-Only Tools (1)
- `POST /v1/tools/developability_report` — assembles an informational developability-signal report from a perturbation result (compound SMILES + observed cellular effect). Chains ADMET + FAVES V4 + clinical clearance + optional docking. Returns structured signals with per-stage provenance and competence gating (clinical signal suppressed for out-of-domain therapeutic areas — oncology / CNS / infectious — rather than emitting a low-confidence number). Marked `x-mcp-exposed: false` in the OpenAPI spec; same request/response envelope as every catalog tool.

## Capabilities

### Data Layer
- Molecular property calculation (LogP, TPSA, HBA/HBD, MW, ~20 more) via RDKit in-process
- 31 ML ADMET prediction models (absorption, metabolism, toxicity, CYP substrates) — served via addie-models when wired
- BOILED-Egg pharmacokinetic classification: GI absorption + BBB permeation
- Local alert screening in the workbench (PAINS + Lipinski + Ghose + Veber + Egan + Muegge + controlled-substance flags)
- FAVES V4 regulatory compliance is a hosted API (not shipped with the OSS engine): DEA, FDA, EPA, CWC, EU REACH, BTWC, Australia, OPCW; 1,585 SMARTS patterns

### Omics Intelligence (optional data pack, on roadmap for OSS v1.1.x)
- 108,436 target-disease associations (Open Targets + expression data)
- 56 pharmacogene documents for patient stratification
- 134,940 ClinVar pathogenic variants (13,252 affecting binding sites)

### Research Search
- Literature (PubMed), patents (Google Patents), preprints (bioRxiv / medRxiv / arXiv), clinical trials (ClinicalTrials.gov), bioactive compounds (ChEMBL)

### GPU Compute Pipeline
- Lead optimization: scaffold hopping patterns with Tanimoto similarity, patent risk
- Molecular docking: AutoDock-GPU with PLIP binding pose analysis, configurable protonation pH
- Molecular dynamics: GROMACS with MM-GBSA, membrane pipeline (CHARMM36m)
- Protein structure prediction: OpenFold3
- Conformational dynamics: AlphaFlow / ESMFlow ensemble generation

### Quantum Chemistry & Neural Potentials
- Semi-empirical QM: GFN2-xTB (energy, optimization, solvation, Hessian/frequency, spin states)
- Conformer search: CREST 3.0.2 (iMTD-GC algorithm)
- Neural network potentials: ANI-2x (organic), MACE-MP-0 (universal), ASE BFGS optimizer
- NEB transition state search: CI-NEB for activation barriers
- Open service wrappers: the GPU compute services ship as standalone open repos (Apache-2.0), each deployable on its own — gromacs-md (MD / GROMACS, https://github.com/NovoMCP/gromacs-md), novomcp-nnp (neural-network potentials, https://github.com/NovoMCP/novomcp-nnp), novomcp-qm (quantum chemistry / xTB / CREST, https://github.com/NovoMCP/novomcp-qm)

### Property Prediction
- pKa: Chemprop v9 (SAMPL8 RMSE 1.255, R² 0.878)
- Solubility: Chemprop D-MPNN with temperature dependence (273–368K)
- Bond dissociation energy: alfabet model for metabolic soft spots

### Clinical Outcome Prediction
- NovoExpert-3: Phase I clearance probability (AUROC 0.756, 63 features, SHAP)
- Competence-gated: cardiovascular / GI validated; oncology / CNS / infectious explicitly out-of-domain

### Materials Science
- Electrolyte redox screening: per-class calibrated potentials
- OLED frontier orbital analysis: sTDA-xTB excited states, motif detection
- Reaction thermodynamics: ΔG, ΔH, TΔS, K_eq with confidence tiering
- Transition state search: CI-NEB activation barriers
- Materials Project: 150K+ inorganic solids

### Enterprise Data Connectors
- Snowflake, Databricks, BigQuery, Supabase
- Bidirectional: pull, enrich, push
- 21 CFR Part 11 audit trail (cloud deployment)

## Governance & Audit (built into the engine)

- **Immutable audit trail** — every AI decision written per-stage across the discovery funnel; reproducible, reconstructable, exportable
- **FAVES V4 compliance** (hosted) — 1,585 SMARTS across 8 jurisdictions, run inline at the point of decision
- **Cross-run memory** — funnel context persists across sessions
- **Agent governance** — RFC 8693 token exchange, per-agent key scoping, agent-level audit on cloud deployments

## Pricing

Two ways to run the engine.

### Self-hosted — Free forever
Run it on your own hardware. You get:
- The complete engine — MCP + REST server, 69 tools
- All ADMET / property prediction models
- All compute-service reference images (docking, MD, QM, NNP)
- Pluggable auth, credit metering, audit sink, LLM provider
- Docker Compose one-liner for a full local stack
- Weekly releases, GitHub issues + Discussions
- License: Apache-2.0 (orchestration core BSL 1.1 → Apache 2029-07-12)

### Cloud deployment — Contact for pricing
NovoMCP handles setup, backups, upgrades. You get:
- Everything in Self-hosted, run by NovoMCP
- Dedicated subdomain (`yourorg.novomcp.com`) + SAML SSO
- FAVES API — SLA, immutable audit-log retention, IQ/OQ/PQ documentation
- GPU compute pool for docking + MD (no queue management)
- Upgrades on the weekly release train, transparent changelog
- Backups + disaster-recovery + regional data residency
- Named on-call engineer + response-time SLA

Contact: https://novomcp.com/sales-inquiry

## Research

Published papers and preprints from NovoMCP (A. Harrison, 2026). Full list at https://novomcp.com/research and Google Scholar (https://scholar.google.com/citations?user=O3Bdfh8AAAAJ):

- FAVES V4: A Production-Deployed Compliance and Liability Screening System for Drug Discovery, Aligned to HHS/ONC FAVES and NIST AI 600-1
- NovoExpert-2: State-of-the-Art ADMET Prediction via Gradient-Boosted Trees on MapLight Fingerprints and GIN Embeddings
- From 2D Graphs to Sub-Unit pKa: Nine Iterations to Beat MolGpKa on Both Scaffolds and Sulfonamides
- Routed Ensemble pKa Prediction with Functional-Group-Specific Atom-Level Descriptors: A Systematic Ablation and Competitive Analysis
- Protocol-Mediated Execution Governance for LLM Agent Systems (SSRN 6426760)
- Molecular Targets Shape Reasoning Topology: Bidirectional Validation of Chemistry-Inspired Bond Classification for AI Reasoning Chains
- FAVES V3: A Three-Tier Regulatory Compliance Detection System for Molecular Databases

## Quick Start

Self-hosted (open source, Apache-2.0):

1. `git clone https://github.com/NovoMCP/novomcp && cd novomcp/orchestrator && pip install -r requirements.txt && python main_https.py` (Python 3.10+)
2. Or `git clone https://github.com/NovoMCP/novomcp && cd novomcp && docker compose up`
3. Point your assistant at the local engine and start calling tools.

Add to an AI assistant:
- Claude Desktop: Settings > Connectors > Add your engine URL
- Cursor / Windsurf / VS Code / Continue.dev / Cody / OpenCode: MCP config with server URL + Bearer key
- ChatGPT / Gemini / Grok / Mistral / Cohere: Settings > MCP / Tools > Add server URL
- Ollama / Open WebUI / NVIDIA NeMo Agent Toolkit: bearer token (API key) over HTTP

Cloud deployment (NovoMCP runs it for you): contact https://novomcp.com/sales-inquiry.

## Drug Discovery — 11-stage funnel (stage pages)

The autonomous discovery funnel is 11 stages in the open engine (FEP is available as a paid Novo Compute service, not part of the OSS funnel):

1. Target Discovery: https://novomcp.com/target-discovery
2. Target Validation: https://novomcp.com/target-validation
3. Literature & Landscape: https://novomcp.com/literature-landscape
4. Known Actives: https://novomcp.com/known-actives
5. ADMET & Properties: https://novomcp.com/engine
6. Compliance: https://novomcp.com/compliance
7. Lead Optimization: https://novomcp.com/lead-optimization
8. Molecular Docking: https://novomcp.com/molecular-docking
9. Clinical Outcomes: https://novomcp.com/clinical-outcomes
10. MD Simulation: https://novomcp.com/molecular-dynamics
11. Patient Stratification: https://novomcp.com/patient-stratification

FEP (available on Novo Compute): https://novomcp.com/fep

## Materials Science — workflow pages

A toolkit, not a funnel:

- Overview: https://novomcp.com/materials-science
- OLED & Emitter Screening: https://novomcp.com/materials-science/oled-screening
- Electrolyte Stability: https://novomcp.com/materials-science/electrolyte-stability
- Reaction Feasibility: https://novomcp.com/materials-science/reaction-feasibility
- Inorganic Discovery: https://novomcp.com/materials-science/inorganic-discovery

## Industries — audience entry points

- Therapeutics: https://novomcp.com/therapeutics
- Biotech: https://novomcp.com/biotech
- Advanced Materials: https://novomcp.com/advanced-materials
- Energy: https://novomcp.com/energy

## Links

- Company: https://novomcp.com
- The Engine: https://novomcp.com/engine
- Drug Discovery: https://novomcp.com/drug-discovery
- Materials Science: https://novomcp.com/materials-science
- Novo Compute: https://novomcp.com/novo-compute
- NovoWorkbench: https://novomcp.com/novoworkbench
- Extensions: https://novomcp.com/extensions
- REST API: https://novomcp.com/api
- FAVES API: https://novomcp.com/faves
- Pricing: https://novomcp.com/pricing
- Documentation: https://docs.novomcp.com/
- Tutorials: https://novomcp.com/tutorials
- Research: https://novomcp.com/research
- News: https://novomcp.com/news
- About: https://novomcp.com/about
- Team: https://novomcp.com/team
- Security: https://novomcp.com/security
- Privacy: https://novomcp.com/privacy
- Terms: https://novomcp.com/terms
- Sales / cloud deployment / FAVES access: https://novomcp.com/sales-inquiry
- GitHub: https://github.com/NovoMCP/novomcp
- Contact: hello@novomcp.com