The ADMETlab MCP server has 4 well-defined tools with consistent naming (verb_noun pattern: wash_, render_, predict_, fetch_), clear descriptions (90-160 chars), and complete JSON schemas with typed parameters. However, parameter descriptions are sparse or missing, output schemas are undocumented, and error handling lacks recovery guidance. Tool composition is sound (each does one thing), but the server does not match production-grade definition quality due to incomplete parameter documentation and missing output specifications.

• Add parameter descriptions to every parameter. Example for 'figsize': 'Optional 2-element array [width, height] in pixels. Default is [300, 300]. Example: [400, 400].' Current schema has no description for figsize.
• Document output schemas in tool descriptions. Example: 'wash_molecule returns {"SMILES": [string], "cleaned_smiles": [string]} containing original and standardized SMILES strings.'
• Explain error recovery in tool descriptions. Example for predict_admet: 'If the upstream service is unavailable (502/503), the server automatically retries up to 3 times with exponential backoff. If all retries fail, an error is returned with guidance to retry manually after a delay.'
• Remove the deprecated 'uncertain' parameter from predict_admet entirely. If backward compatibility is needed, note this in the source code comment rather than exposing it to LLM callers.
• Add dependency hints to descriptions. Example: 'Call wash_molecule first to standardize SMILES strings before prediction for better accuracy. predict_admet accepts both washed and unwashed SMILES.'
• Specify result limits in descriptions. Example: 'Accepts up to 1000 SMILES per call (configurable via ADMETLAB_BATCH_SIZE). For larger batches, split into multiple calls.'
• Document the taskId format for fetch_admet_csv. Example: 'The taskId returned by predict_admet when processing completes asynchronously. Format: UUID string.'
• Add constraints to numeric parameters. Example for figsize: 'Each dimension must be between 100 and 2000 pixels.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry