Claude security directory

Inductive Bio

Published by Inductive Bio

High risk

Inductive Bio’s absorption, distribution, metabolism, excretion, and toxicity (ADMET) models predict the chemical and pharmacokinetic properties that determine whether a molecule can become a viable drug. When these predictions are surfaced directly in conversation, scientists can evaluate compounds, triage ideas, and prioritize which molecules to synthesize, without leaving their existing workflow. Through this MCP server, users can access Inductive Bio’s models for the physicochemical properties LogD (lipophilicity) and pKa (acid/base ionization), key drivers of solubility and permeability. Inductive’s full suite of models spans all tiers of ADMET assays, including microsomal stability, efflux, CYP inhibition, brain penetration, hERG inhibition, and more. Models are fine-tuned to each customer’s chemical space to advance predictive performance. To access the full suite, reach out to Inductive Bio here: https://www.inductive.bio/book-a-demo.

Security assessment

Plutonium assessed this web connector for permissions, capabilities, and security-relevant behavior.

  • Reads your private information: Property prediction requests may include proprietary chemical structures (e.g., SMILES) and related context, which can constitute sensitive IP even if the connector is nominally read-only.
  • Sends your data to outside companies: Inputs sent to Inductive Bio’s hosted service for prediction are transmitted to an external vendor, which may be subject to their retention/logging and privacy practices.
  • Can run commands or queries on your behalf: The prediction tool effectively triggers remote model execution (server-side inference) on user-supplied inputs; while not local code execution, it performs compute/actions on your behalf and could be abused for high-volume/automated queries.
  • Stores long-lived access tokens: Access to a hosted MCP server typically involves API keys or tokens; if long-lived credentials are used, compromise could allow continued access to query models and submit sensitive structures.

Available capabilities

This add-on exposes 2 tools or capabilities.

  • list_available_models
  • predict_properties

The interactive security report will load automatically.