sciagentkit 1.0.3

Deterministic scientific tools for AI agents — real computational biology & drug discovery, no invented numbers.

Website · Pricing · PyPI · Support

SciAgentKit gives Claude, Cursor, Gemini CLI and Codex deterministic tools for drug discovery — molecule & ADME/tox analysis, target/PDB selection, docking, molecular dynamics, target-aware hit ranking, wet-lab plans, and signed reports — instead of letting the model invent descriptors, docking scores, or RMSD values.

Skills decide the workflow. Deterministic tools run the science. Reports preserve the evidence.

Install

Full pipeline (recommended) — one conda environment with every open-source engine (RDKit, OpenMM, AutoDock Vina, Open Babel, fpocket, …) plus the toolkit:

curl -O https://sciagentkit.com/environment.yml
conda env create -f environment.yml
conda activate sciagentkit

Analysis only — pip (molecule/ADME/tox/scaffolds/target selection; no docking/MD):

pip install "sciagentkit[mcp]"

Get your API key at https://sciagentkit.com (free 3-day trial), then:

export SCIAGENTKIT_API_KEY=sak_your_key      # Windows: setx SCIAGENTKIT_API_KEY "sak_..."

Use it in your AI agent

Claude Code

claude mcp add sciagentkit -s user \
  --env SCIAGENTKIT_API_KEY=sak_your_key \
  -- python -m sciagentkit.servers.rdkit_server

Cursor / Gemini CLI / Codex

sciagent init-agent cursor       # or: gemini | codex | all

Then ask your agent naturally:

"Dock these ligands against EGFR L858R, rank target-aware top hits, and write a report."

Full step-by-step guide (all four agents): see USER_GUIDE.md or sciagentkit.com.

What it does

• Molecule libraries — canonicalization, descriptors (MW, logP, QED, TPSA), scaffold diversity, novelty audits (RDKit)
• ADME & toxicity — ESOL solubility, drug-likeness (Lipinski/Veber/Ghose/Egan), GI/BBB heuristics, PAINS/Brenk alerts
• Targets — UniProt resolution + mutation-aware PDB ranking
• Docking — pocket detection (crystal-ligand / fpocket / P2Rank) + Vina/GNINA/smina, ranked poses
• Molecular dynamics — OpenMM setup + trajectory RMSD/RMSF + ProLIF interactions
• Target-aware hit ranking — combines docking, MD stability, and ADME/tox (toxicity penalized; BBB up-weighted for CNS targets)
• Wet-lab validation plans — staged assays + recommended cell lines for the target
• Signed, reproducible reports — Markdown / PDF / DOCX with a run_manifest.json (input hashes + tool versions)

Every value is computed by real tools, never inferred by a language model. Outputs are saved under runs/ with a reproducibility manifest.

Free vs paid

Team plans are per seat. Manage your plan at sciagentkit.com.

Scientific limits

SciAgentKit is a workflow and tooling layer, not expert review. Docking scores are ranking heuristics, not binding free energies; short MD validates setup and analysis plumbing, not equilibrium stability. Receptor protonation, missing loops, cofactors, metals, membrane context, and ligand parameterization still require expert review.

Support

Account, docs, and billing: https://sciagentkit.com — questions: support@sciagentkit.com

© 2026 Istanbulmedchem. All rights reserved.