Claude 现已支持 Gmail 邮件与 Google Drive 文件管理

Core Highlights

Anthropic's Claude gained two new connectors this week: Gmail and Google Drive. In plain terms, Claude can now step directly into your inbox to send messages and into your cloud drive to manage files, turning the chatbot into an office assistant that can actually take action. The feature is available to all paid plan users, and you only need to pick the service from the connector menu and finish authorization to get started, so the barrier to entry is low. The addition signals a clear shift from purely conversational AI toward tool-using agents that operate inside the apps people already use every day. It also shows how quickly the assistant category is moving from answering questions to completing real tasks on a user's behalf within trusted accounts. For Anthropic, the connector model is becoming the main way it competes on usefulness without having to rebuild the underlying productivity suite itself.

What Happened

The usage is direct: you ask Claude to reply to an email thread, and it first drafts the reply and then sends it for you. You can also tell it to create, organize, or search documents inside Google Drive. The key point is that permissions stay under your control: the system tells you when your own approval is required, and sensitive actions will not run automatically without confirmation. The company announced this in a post on the X platform and stressed that you control when approval is needed. This design of acting only after human review lets automation and reassurance coexist instead of conflicting, which is exactly what nervous IT departments want to hear before they allow any AI near employee mailboxes. The post framed the launch as a direct response to user demand for Claude to act inside everyday tools, not merely discuss them.

Technical Details

A connector is the bridge through which Claude reaches external applications. The user must explicitly authorize it from the menu, and the credentials are held by the corresponding service provider rather than being baked into model weights. Both email and file access are limited by scopes, so Claude can only touch the account content you have explicitly connected and authorized, and cannot reach beyond into other data. This on-demand, least-privilege design continues Anthropic's consistent philosophy around controllable agents, and it also answers the compliance concerns enterprises care about most when granting AI access to production systems. Scope limits mean a mistake or a prompt injection cannot silently exfiltrate an entire workspace. Because tokens and credentials never enter the model weights, a compromised session stays contained to the granted scope rather than exposing the assistant's internals.

Compared to Competitors

Compared with ChatGPT's Connectors and the deep suite binding of Microsoft 365 Copilot, Claude took a lightweight, cross-service route: instead of locking into one office ecosystem, it plugs in one independent connector at a time. This is especially friendly to Google-centric users who run both Gmail and Drive, and it directly fills the gap Claude had in email automation. Its differentiation is not feature stacking but returning control to the user, which can matter more in real workflows than a longer list of integrations that the user cannot govern. By staying modular, Claude also avoids forcing customers to commit to a single vendor's productivity bundle.

Industry Impact

For knowledge workers, this means routine email archiving, follow-ups, and document tidying can be partially automated, freeing people from repetitive labor. In enterprise settings, assistant-style AI connecting to production-grade SaaS is becoming a trend, yet approval gates and least privilege remain the keys to safe deployment. Claude's step pushes the AI that can chat further toward an AI that can act, and it lays groundwork for connecting more office applications later without rewiring the trust model that enterprises depend on. The direction is clear: the next battleground for assistants is not who writes the best paragraph, but who can safely do the most useful work inside the tools teams already trust. As more connectors arrive, the competitive question shifts from model quality alone to the richness and safety of the action layer wrapped around the model. Early users report that draft quality is high enough that most replies need only a quick glance before approval, which keeps the human in the loop without slowing them down.