NovoMCP

Overview

NovoMCP is a computational chemistry engine intended for drug-discovery and materials-science workflows. Its MCP connector and REST API let clients request chemistry operations from a shared backend, while CLI and visual interfaces provide alternative access. The engine wraps scientific packages and services rather than replacing them: the README names RDKit, GROMACS, AutoDock-GPU, OpenFold, Chai, Boltz, Gnina, xTB, ANI-2x, AIMNet2, and MACE among its dependencies or integrations.

A documented starting workflow submits a SMILES string to get_molecule_profile. The described response includes molecular properties, structural alerts, and ADMET predictions when the relevant prediction service is connected. Local property calculations use RDKit; ADMET depends on wiring the addie-models service. Beyond profiling, the source describes literature and database searches, library screening, and optional docking, molecular dynamics, quantum calculations, and protein-structure prediction. A staged discovery funnel coordinates tool execution, with each call carrying a funnel_id and reaching an audit sink.

Availability depends on deployment configuration. A bare local installation exposes a subset of tools, while additional capabilities require optional services; the dashboard reports their status and configuration requirements. Local defaults omit authentication and credit accounting and write audit records to a file. Hosted deployments use separately configured authentication, metering, and auditing implementations. The excerpts provide setup and architecture descriptions, not scientific validation results. Repository code is available under split terms: most wrappers and interfaces use Apache-2.0, whereas the orchestration core currently uses Business Source License 1.1 with a specified future conversion.

Key Features

• MCP access for compatible clients and a curated REST interface using `POST /v1/tools/{name}`, with an OpenAPI description at `/v1/openapi.json`.
• SMILES-based molecular profiling through `get_molecule_profile`, with local RDKit properties and service-dependent ADMET predictions.
• Documented search tools for ChEMBL, clinical trials, bioRxiv, and PubMed literature, alongside library screening and molecular-dynamics pre-flight checks.
• Optional service wrappers for docking, molecular dynamics, quantum calculations, neural-network potentials, and protein-structure prediction.
• Discovery-funnel orchestration with `funnel_id` tagging and configurable audit sinks; local auditing writes JSON-lines to `~/.novo/audit.jsonl`.
• A Next.js dashboard for molecule profiles, engine and service status, and configuration of keys, compliance, observability, and data connectors.

Use Cases

• Suggested evaluation: submit known molecular SMILES to inspect property and structural-alert outputs, then assess separately configured ADMET predictions.
• Suggested evaluation: combine literature and ChEMBL searches with library screening to support early candidate triage.
• Suggested evaluation: connect docking or molecular-dynamics services and examine how their outputs fit into a staged discovery workflow.
• Suggested evaluation: inspect audit records to assess traceability of chemistry tool calls across an MCP-driven workflow.

How to Use

1. Read the official README and choose local or hosted operation. It specifies Python 3.10–3.12 for the backend; treat this as a documented requirement, not independently tested compatibility.
2. Obtain the code from the NovoMCP repository. Follow the README’s virtual-environment, dependency-installation, and main_https.py startup procedure. Its clone example contains an organization placeholder, so use the actual repository link.
3. Follow the README’s get_molecule_profile example with its supplied aspirin SMILES. Inspect properties and alerts, and check whether the ADMET service is connected before expecting prediction output.
4. Select MCP, REST, or the dashboard as your interface. Consult the README’s dashboard instructions and capability display before configuring optional scientific services.
5. Evaluate outputs against appropriate reference cases and inspect local audit records. Before deployment or redistribution, read both the top-level license and core license, since their scopes and conditions differ.