The compute engine behind the discovery
GPU docking, molecular dynamics, quantum chemistry, neural network potentials, structure prediction, and materials science. The simulation and prediction layer that powers the pipeline — accessible from any MCP-compatible AI assistant at compute.novomcp.com.
What Novo Compute adds
Everything that requires GPU, QM, or neural network inference. Requires a paid tier (Core, Scale, or Enterprise).
GPU-accelerated docking
AutoDock-GPU with reference ligand co-docking, PLIP contact analysis, strain energy validation. 3-10 seconds per molecule.
Molecular dynamics
GROMACS on GPU. Soluble and membrane proteins. MM-GBSA binding free energy. RMSD/RMSF/H-bond analysis. Async with email notifications.
Quantum chemistry
GFN2-xTB semi-empirical QM. Energy, optimization, solvation, Hessian/frequency, spin states. CREST conformer search.
Neural network potentials
ANI-2x (organic) and MACE-MP-0 (universal). ~100x faster than xTB. Geometry optimization in milliseconds.
Structure prediction
OpenFold3 de novo prediction. AlphaFlow conformational dynamics. Smart PDB resolver for experimental structures.
Property prediction
Custom pKa (Chemprop), solubility with temperature dependence, bond dissociation energy for metabolic soft spots.
Materials science
OLED screening, electrolyte stability, reaction thermodynamics, transition state search, Materials Project integration.
Metalloprotein support
MCPB.py QM-to-force-field bridge. audit_system pre-flight classification routes complex proteins automatically.
Tool inventory
27 tools across 6 domains. All on Novo Compute (compute.novomcp.com).
GPU Compute
dock_moleculesAutoDock-GPU with reference ligand co-docking, PLIP contact analysis, configurable protonation pH. Two-phase: estimate → execute.
run_molecular_dynamicsGROMACS 2023.3 on GPU. Soluble (AMBER99SB-ILDN) and membrane (CHARMM36m via PACKMOL-Memgen). MM-GBSA binding free energy.
parameterize_metalQM-to-force-field bridge via MCPB.py. Gaussian/ORCA QM logs → AMBER .frcmod/.prep + GROMACS .top/.gro with RESP charges.
audit_systemFree pre-flight protein classification. Membrane detection (OPM), metal sites, heme/Fe-S clusters. Routes complex cases before committing credits.
Structure Prediction
get_protein_structureSmart resolver: gene name or PDB ID → experimental structure from RCSB with metadata, chains, ligands. Inline PDB data.
predict_structureDe novo protein structure prediction via OpenFold3. Per-residue pLDDT confidence scores.
generate_dynamicsAlphaFlow/ESMFlow conformational dynamics ensemble. Multi-model PDB with per-residue RMSF and PCA.
Quantum Chemistry
run_qm_calculationGFN2-xTB: single-point energy, geometry optimization, solvation (ALPB), Hessian/frequency, spin states.
run_conformer_searchCREST 3.0.2 (iMTD-GC). Ranked conformers with Boltzmann populations. Essential before docking.
dock_with_strainStrain energy of docked poses via GFN2-xTB. High strain (>5 kcal/mol) flags artifact poses.
run_qm_hessianVibrational frequencies, ZPE, Gibbs corrections. Confirms transition states or true minima.
Neural Network Potentials
compute_energyANI-2x (organic) or MACE-MP-0 (universal). ~100x faster than xTB. Energy in eV and kcal/mol + atomic forces.
optimize_geometry_nnpASE BFGS geometry optimization. ~10x faster than xTB. Ibuprofen: 11 steps, 373ms.
Property Prediction
predict_pkaCustom Chemprop D-MPNN. pKa values, ionizable groups, confidence. SAMPL8 RMSE 1.255, R² 0.878.
predict_solubilityAqueous solubility (LogS) with temperature dependence (273-368K). Returns mg/mL and category.
predict_bdeBond dissociation energies via ALFABET. Metabolic soft spots (BDE < 85 kcal/mol).
Materials Science
predict_frontier_orbitalsHOMO, LUMO, gap, S1/T1 energies, device role. 14 OLED motifs detected automatically.
run_excited_statesFull singlet/triplet ladder via sTDA-xTB. Oscillator strengths for TADF screening.
predict_redox_potentialOxidation/reduction potentials. Per-class calibration. 5 reference electrodes. 4 voltage windows.
predict_reaction_thermodynamicsΔG, ΔH, TΔS, K_eq. Confidence-tiered for organic, mixed, and metal-containing systems.
find_transition_stateCI-NEB activation barriers. Forward/reverse barriers. TS geometry and minimum energy path.
search_materials_project150K+ inorganic materials. Band gaps, formation energies, stability, space groups.
How to connect
Novo Compute is a separate MCP server at compute.novomcp.com/mcp. Add it alongside Novo in your AI assistant settings. Both servers share the same credit pool — no separate billing.
Requires a paid tier (Core credit pack, Scale, or Enterprise). Free trial users see an upsell message with a link to upgrade.
Works with any MCP client
Claude. ChatGPT. Cursor. Windsurf. GitHub Copilot. VS Code. One URL, one API key, and every compute tool appears in your conversation — without local installation or GPU provisioning.
One-click setup
Add compute.novomcp.com, authenticate once. OAuth handles the rest. New tools appear server-side.
Natural language
Describe what you need. Your AI selects the right compute tool — docking, MD, QM, or NNP — and runs it on GPU.
Always current
New force fields, new models, new compute capabilities — deployed server-side without reconfiguration.
The compute layer your AI is missing
GPU docking. Molecular dynamics. Quantum chemistry. Sign up and run your first simulation.