Skip to content

Use case: AI-assisted molecular analysis with MCP

Problem

You want an AI agent (Claude, GPT-4, etc.) to reason about molecular structures — but LLMs have no native chemistry tools. You need a bridge that lets the AI call real cheminformatics functions on actual SMILES strings.

Solution

chematic ships a built-in MCP server (chematic-mcp) that exposes chemistry tools to any MCP-compatible AI agent. Wire it to Claude Desktop once; the AI can then evaluate, search, and synthesise molecules in natural conversation — no Python environment on the client side.

Setup

# Install chematic Python package (no C/C++ toolchain needed)
pip install chematic

# Or build the MCP server binary from source
cargo build -p chematic-mcp --release

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "chematic": {
      "command": "/path/to/chematic-mcp"
    }
  }
}

What the agent can do

Once connected, Claude Desktop can call these tools:

Tool What it does
evaluate_molecule MW, LogP, TPSA, ADMET, PAINS, Lipinski
similarity_search Find similar structures in a SMILES list
substructure_search SMARTS-based filter across a set
generate_3d ETKDG 3D coordinates + MMFF94 minimization
retro_disconnect Retrosynthesis (60 reaction templates)
standardize Salt stripping, neutralization, tautomer canon.

Example conversation

User: I have a hit compound CC(=O)Nc1ccc(O)cc1. Is it drug-like?

chematic evaluates the molecule and returns:

Molecular weight 151.2 Da, formula C8H9NO2.
LogP 0.46 (mildly lipophilic), TPSA 49.3 Ų.
HBD 2, HBA 3, 2 rotatable bonds, 1 aromatic ring.
Drug-likeness: no Lipinski rule-of-5 violations. Likely orally bioavailable.
QED 0.67. No structural alerts (PAINS / Brenk clean).

User: Find me 5 analogs from this SMILES list that are most similar.

The agent calls similarity_search and returns ranked results with Tanimoto scores.

Scripting with Python

import chematic

mol = chematic.from_smiles("CC(=O)Nc1ccc(O)cc1")  # paracetamol

# Natural-language summary (same as what the MCP tool returns)
print(mol.describe())

# Full ADMET profile
print(mol.admet())

# Retrosynthesis
routes = mol.retro_disconnect(max_results=5)
for r in routes:
    print(r)

Typical agentic workflow

User query
    ↓
Claude reads SMILES / name
    ↓
chematic MCP: standardize → evaluate → similarity_search
    ↓
Claude interprets results and suggests next experiment
    ↓
chematic MCP: retro_disconnect (if synthesis needed)
    ↓
Claude writes experimental plan

chematic is the chemistry engine; the AI handles language, reasoning, and decision-making. No Python environment is needed on the client — the MCP server binary handles everything.

  • mol.describe() — natural-language property summary (the same text the MCP server returns)
  • mol.admet() — BBB, Caco-2, hERG, CYP3A4 in one call
  • mol.retro_disconnect() — retrosynthesis with 60 reaction templates
  • chematic-mcp binary — run cargo build -p chematic-mcp --release