Anthropic Expands Claude Connections to Personal Apps

Anthropic is expanding Claude with a new set of consumer-focused connectors, letting the assistant access and act on data inside apps such as Spotify, Uber Eats, Intuit TurboTax, AllTrails, Instacart, Booking.com, Resy, and more. The company says the connector catalog has grown to 200+ connectors since July 2025 and that the personal-app additions are available to all Claude plans, with mobile support in beta.

The connectors operate as OAuth-style integrations that grant Claude scoped access to a user's app data and actions. Claude now surfaces connectors dynamically in a conversation based on intent and context, and when multiple connectors could help, it presents both and ranks them by perceived usefulness. Anthropic enforces an explicit confirmation step for high-impact actions such as purchases or reservations. Key product behaviors and safeguards highlighted by Anthropic include:

• •Ranked suggestions of relevant connectors inside the chat, driven by conversation context and user preferences
• •Explicit confirmation before executing transactional actions against a connected service
• •User controls to add or remove connectors from the Claude sidebar
• •Privacy commitments: data pulled from a connected app is not used to train models and apps do not see other Claude conversations

Consumer-facing connectors are now a standard capability among major assistant platforms; competitors have already exposed similar integrations in recent product cycles. What differentiates this update is the dynamic in-chat suggestion model, combined with Anthropic's emphasis on noncommercial ranking and a declared ad-free experience. For practitioners, this means more end-to-end workflows can be composed inside a single conversational session: itinerary planning that queries TripAdvisor and books with Booking.com, grocery list creation that pushes items to Instacart, or tax preparation assistance that pulls records from Intuit TurboTax.

granting conversation-driven access to personal apps widens the permission surface. The useful safeguards Anthropic lists are necessary but not sufficient for production risk management. Teams building or operating similar integrations should verify token lifecycles, consent scopes, replay protection, and audit logging. From an ML perspective, the claim that connector data is not used for model training reduces one category of data governance risk, but operators still need to protect connectors from unintended data exfiltration and ensure downstream services cannot infer sensitive context from actuation history.

Adoption patterns and abuse vectors. Track whether dynamic in-chat suggestions increase accidental transactions, and whether third-party apps object to ranking or data usage. For product teams, confirm how revocation and consent UI behave across platforms and how the system logs actions for forensics and debugging.

implement least-privilege OAuth scopes, short-lived tokens, deterministic confirmation UX for transactions, and per-connector telemetry to detect anomalous calls. Expect a wave of product work to standardize connector SDKs, permission models, and rigorous security reviews as more consumer apps open programmatic access to assistants.