Preface

On August 3, 2026, Cursor released the Google Workspace plugin in its official Changelog. Developers can now let their coding Agent directly access Gmail, Google Drive, and Google Calendar within the IDE, search emails, read and write files, and manage schedules without frequently switching to a browser.

This integration was not built by Cursor itself, but was completed through the official Google-hosted remote MCP (Model Context Protocol) server. The Agent calls standard tools provided by Google under *.googleapis.com/mcp/v1, with authorization via OAuth 2.0, and permission boundaries consistent with the user’s Google account. For developers who code in Cursor every day and heavily rely on Google collaboration tools, this is a very practical expansion in the Marketplace ecosystem; it also brought issues such as “whether the Agent can read enterprise emails or modify cloud documents” to the forefront.

This article cross-verifies based on Cursor’s official Changelog, the Cursor Marketplace plugin page, and Google Workspace developer documentation, sorting out the capability boundaries, installation methods, and security considerations of this update.

Background: MCP and Google Remote Servers

MCP is an open standard that allows AI applications to call external tools and data through a unified protocol. Cursor has continuously expanded its MCP and Plugins capabilities since 2025; Google opened the Workspace MCP server to Public Developer Preview during Cloud Next in May 2026.

Google provides remote MCP endpoints for each Workspace product, such as:
- Gmail: https://gmailmcp.googleapis.com/mcp/v1
- Google Drive: https://drivemcp.googleapis.com/mcp/v1
- Google Calendar: https://calendarmcp.googleapis.com/mcp/v1

The three official Cursor plugins in the Marketplace are labeled Created by Cursor · Verified by Cursor, and their descriptions all state that they connect to the corresponding services through Google’s remote MCP server. That is to say: the plugin is responsible for completing OAuth and tool registration on the Cursor side, and the actual read and write actions are executed by the MCP service hosted by Google, which inherits the Workspace permissions of the currently logged-in user.

It should be noted that Google Workspace MCP is still in the developer preview stage, and the tool list and quotas may change with version updates. The functional descriptions below are subject to Cursor’s official release notes; if you care about a specific operation (such as “whether the Agent can send emails directly”), it is recommended to check the tools/list exposed by the MCP in the current session after connecting, and refer to the real-time capabilities.

What the Three Plugins Can Do

Cursor launched three independent plugins this time, which can be installed as needed.

1. Gmail

The capabilities listed in the official Changelog include:
- Search and read emails
- Draft and send messages
- Apply labels and manage email threads

The description on Cursor Marketplace is: search, read, draft, label, and manage email. Typical tools in the Google Gmail MCP developer documentation include gmail.search_threads, gmail.get_thread, gmail.create_draft, gmail.label_thread, etc.; the documentation examples still suggest that users review and then send emails in the Gmail interface after creating a draft. Given that MCP is in the preview period, whether the sending capability in “draft and send” is fully open to the Agent shall be subject to the actual tool list in your current environment.

2. Google Drive

Official capability overview:
- Search files and folders
- Open and download content
- Create and organize files

The Marketplace description also mentions share and manage files. Google Drive MCP supports operations such as retrieval, reading metadata and content, and creating files; specific write permissions and sharing tools may also be adjusted during the preview iteration.

3. Google Calendar

Official capability overview:
- Read schedules
- Create and update events
- Find free time

The tools listed in the Google Calendar MCP documentation include list_calendars, list_events, create_event, update_event, suggest_time, etc., which are suitable for letting the Agent help you check files and schedule review meetings while coding.

Installation and OAuth Authorization

There are two installation paths, consistent with the official instructions:
1. Open the Customize page in the Cursor sidebar, and add the corresponding plugin in the MCP / Plugins area;
2. Visit Cursor Marketplace, search for the official Gmail, Google Drive, and Google Calendar plugins and install them.

The Marketplace also provides shortcut commands, such as /add-plugingmail, /add-plugingoogle-drive, and /add-plugingoogle-calendar.

When using it for the first time, Cursor will guide you to complete Google account OAuth authorization. This is a standard OAuth 2.0 process: the Scopes you explicitly agree to grant determine which emails, files, and calendars the Agent can access. If you build your own MCP client to connect to the Google remote server, you also need to enable the corresponding API and MCP services (such as gmailmcp.googleapis.com) in the Google Cloud project; when installing through the official Cursor plugin, the OAuth interaction is usually encapsulated by the Cursor side, and ordinary users do not need to manually configure the Cloud project.

It is recommended to access for the first time:
1. Only install the plugins that are truly needed in the current workflow to avoid granting too many permissions at one time;
2. Use a test account or non-sensitive directories/calendars to verify read and write behaviors;
3. Before the Agent performs write operations (creating files, updating schedules, creating email drafts), confirm the tools and target objects it plans to call.

Typical Usage Scenarios

Combined with the official release notes and MCP tool design, the following scenarios are relatively natural:

Pull requirement context from Drive. Let the Agent search for PRDs, design specifications or interface documents in Drive, and include the content into the current coding session, reducing manual copy-pasting.

Draft collaborative documents or emails based on code changes. After completing a round of function development, ask the Agent to create/update a document in Drive or generate a reply draft in Gmail based on the diff or commit description.

Link schedules and development rhythm. Query the free time in Calendar and create code review or standup meetings without leaving the IDE.

The common point of these scenarios is to converge “context acquisition” and “lightweight collaboration actions” into the Agent conversation, instead of switching back and forth between the IDE and browser tabs.

Permissions and Security: Worth Thinking About in Advance

When the Agent can read Gmail and modify Drive files, the risk scope is no longer just “accidentally modifying local code”, but also includes:

OAuth Scope and account boundaries. The Agent can only access the data visible to the authorized account; the policies for personal Gmail and enterprise Google Workspace accounts are different. Enterprise administrators can manage API access in Security > API Controls of the Admin Console (Google’s official blog states for administrators: there are no mandatory operations during the preview period, but API management capabilities already exist).

Write operations and accidental sending. The tool set during the preview period may have differences between “creating drafts” and “sending directly”; for operations such as emails, sharing files externally, and deleting schedules, it is advisable to adopt the strategy of “Agent drafting + manual confirmation”, especially in production data or customer email scenarios.

Prompt injection and indirect instructions. Google’s developer documentation reminds that the body of emails and the content of documents may be constructed as malicious prompts to induce the Agent to perform unexpected operations. Google provides security capabilities such as Model Armor for integrators to refer to; as a user, you should keep the MCP tool calls that the Agent will execute visible and interruptible.

Preview period changes. The tool names, quotas, and GA time of Workspace MCP have not been finally locked. After the plugin or Google side updates, you should recheck the permissions and tool list.

There have been security checklist discussions about the Cursor Google Workspace plugin in the community (such as practical articles on DEV Community), and the core suggestions are consistent: pilot on a small scale, separate read and write verification, and manually review write operations by default.

Summary

The Google Workspace plugin released by Cursor on August 3, 2026, connected Gmail, Drive, and Calendar to the coding Agent within the IDE, relying on Google’s official remote MCP server and OAuth 2.0 at the bottom, which is an important step for Cursor Marketplace in the direction of “development + collaboration” integration.

For developers, the value lies in reducing context switching and allowing the Agent to directly consume cloud collaboration data; for teams, it is necessary to synchronously evaluate the OAuth authorization scope, preview period tool capabilities, and enterprise API management strategies. If you are already using Cursor’s MCP / Plugins system, you might as well start with read-only scenarios (search emails, read Drive documents, list schedules) before gradually enabling write operations such as creating drafts and updating files.

Official references:
- Google Workspace Plugins · Cursor Changelog
- Cursor Changelog
- Gmail / Google Drive / Google Calendar · Cursor Marketplace
- Configure the Google Workspace MCP servers · Google for Developers