ChemCP

Overview

ChemCP connects an AI-assisted conversation to a molecular structure viewer. Its documented purpose is to display 2D diagrams from valid SMILES strings and show associated molecular descriptors without leaving the chat. It is an MCP server and embedded interface built around RDKit.js, rather than a replacement for the upstream cheminformatics toolkit. The README describes setup through the chemcp npm package and recommends Claude Desktop as the host.

The workflow begins when the assistant calls render_molecule with a SMILES string. ChemCP passes that input to its MCP App interface, where browser-loaded RDKit.js parses the structure, produces an SVG diagram, and computes descriptors client-side. Outputs include molecular weight, LogP, hydrogen-bond donor and acceptor counts, TPSA, rotatable bonds, and rings. Users can also enter SMILES directly or select example molecules in the interface. Name-based prompts are documented, but the described rendering tool receives SMILES; the README does not establish a separate chemical-name lookup service.

Interactive display depends on a host with MCP Apps support. Other MCP clients may receive SMILES as text without the viewer. The Claude.ai connector route is explicitly experimental, with content-security restrictions that can prevent RDKit.js from loading. The interface also depends on access to an external CDN. Source and development instructions are available, but the supplied package manifest contains only a placeholder test command, not evidence of scientific validation or an automated test suite.

Key Features

• Renders 2D molecular structure diagrams as SVG from valid SMILES strings using RDKit.js.
• Computes molecular descriptors client-side, including molecular weight, LogP, TPSA, hydrogen-bond counts, rotatable bonds, and rings.
• Provides an embedded interface for entering SMILES and selecting example molecules inside an MCP Apps-enabled conversation.
• Exposes the `render_molecule` tool and serves the accompanying viewer resource through its MCP server.
• Supports a local stdio setup and an HTTP setup; the README documents an experimental remote-connector workflow for Claude.ai.

Use Cases

• Suggested evaluation: use known SMILES examples to inspect how molecular connectivity is displayed during chemistry teaching or discussion.
• Suggested evaluation: render related molecules, such as caffeine and theobromine, to support a visual structure comparison alongside their displayed descriptors.
• Suggested evaluation: assess whether an MCP Apps-enabled assistant can provide useful structure previews within an existing cheminformatics conversation.
• Suggested developer evaluation: use the documented basic-host workflow to investigate viewer loading and host integration before adopting ChemCP.

How to Use

1. Read the ChemCP setup guide and confirm the documented Node.js prerequisite. Check that your chosen host supports MCP Apps; ordinary MCP tool support alone does not provide the interactive viewer.
2. Follow the guide’s installation instructions for the chemcp package. For Claude Desktop, add the documented chemcp entry to the existing mcpServers configuration, then restart the application.
3. Start with an explicit input such as CCO and ask the assistant to render that SMILES. Check whether the structure and descriptor panel appear, rather than treating a text response as a successful viewer test.
4. Try direct SMILES entry and the example molecules in the interface. As an evaluation step, compare the displayed structure and descriptors with an independently trusted reference.
5. If rendering fails, consult the README’s host and CDN troubleshooting sections. Use its experimental HTTP connector instructions only after considering the documented Claude.ai restrictions; developers can instead follow the linked basic-host development route.