Use Cases
Six example workflows. Each runs 1:1 in Claude, ChatGPT or Cursor.
Frontier LLMs reach 14 to 41 percent on chemistry. With CovaSyn MCP attached: 76 to 92 percent.
Peer-reviewed on MolecularIQ (Klambauer Lab JKU, ICLR 2026). 3,540 verified chemistry tasks, four frontier models. Three reach 85–92 %; the cheapest (Gemini 3.5 Flash) lands at 76 %.
Retrosynthesis in chat
Claude analyses the target molecule, proposes routes, checks druglikeness and ADME profiles, and exports the building-block list. Routed through the CovaChem suite.
Plan an ICH stability study
Generate ICH-Q1A-compliant plans, fit Arrhenius models, derive shelf-life bands, straight from a ChatGPT tab. Routed through the CovaStab suite.
Docking screen from Cursor
From a Cursor notebook: load a SMILES list, run a docking pipeline, rank hits by binding energy and synthesisability. Routed through the CovaDock and CovaFold suites.
Formulation optimisation with DoE
Recipe optimisation via RSM designs in Copilot. Proposal, data-collection plan, model fit, robustness check, recommendation, all routed through the CovaDoE and CovaOpt suites.
Antibody stability profile
From a sequence: hydrophobicity patches, aggregation risk, pI, viscosity hotspots. Routed through the CovaBio and CovaStab suites.
Solubility screen for salt selection
pH-dependent solubility curves, antisolvent strategies, mixture predictions for salt form selection. Routed through the CovaSolve suite.
Frequently asked
- Which use case is the best starting point?
- For CDMOs and pharma CMC teams, ICH stability studies are the fastest entry: defined outputs, clear audit requirements, immediately productive. For medicinal chemistry and drug discovery: retrosynthesis or docking. For formulation teams: DoE. Most customers start with one use case and expand horizontally.
- How do I integrate CovaSyn into Claude, ChatGPT or Cursor?
- Add the MCP server entry to your client config, paste your API key from the CovaSyn dashboard. 30 seconds, no code. Drop-in configs for Claude Desktop, Cursor, VS Code and Python are MIT-licensed on GitHub (covasyn-mcp-examples).
- My use case isn't listed, what now?
- These six cases are examples from discovery and CMC. We cover 130+ tools across 8 suites, so usually something fits. Send your workflow via Contact Sales, we come back with concrete tool suggestions.
- How long does a typical tool call take?
- Cheminformatics tools (canonicalize, druglikeness, fingerprints) under 1 second. ADME and tox prediction 2 to 5 seconds. Stability models and Arrhenius fits 5 to 15 seconds. GPU-bound tools like docking and folding 30 seconds to 2 minutes depending on complexity.
- Can IP-sensitive projects run on-premise?
- Yes. On the Enterprise tier we ship CovaSyn as a Docker container for your own infrastructure, single-tenant, no data egress, with a GxP validation pack. Talk to sales for setup.
- What does the audit trail for ICH M7 and ICH Q1 workflows look like?
- Every tool call writes a record: input, tool version, model version, configuration, result, timestamp. Exportable as JSON or CSV for validation reports and inspections. GAMP 5 Software Category 4.
Tool not listed, or could not find a fit?
Tell us briefly what you need. If we have it, we come back with the matching tool. If we do not, we check whether it fits into our MCP, and you hear first.
Email is optional. By entering it you consent to us responding to your wish, the data is used only for that response.