Antigravity for Visual Studio

Bring Google's agent-first development platform to Visual Studio.

Antigravity pairs Google's Gemini models with the Shared Agent Harness to work as an active, context-aware development partner inside Visual Studio, not a passive chat window. It plans and executes multi-step work: it spins up specialized, sandboxed subagents, runs tests, connects to your data through the Model Context Protocol (MCP), and carries out complex refactorings end to end, all under your supervision. Because that work runs in the background, Visual Studio stays responsive, even on large solutions.

Key features

• Autonomous agent orchestration: Delegate complex engineering tasks to a hierarchy of specialized subagents that run in secure, ephemeral sandboxes, so Visual Studio stays responsive.
• Implementation plans: Ask Antigravity to generate a plan before it makes multi-file changes, so you can review the approach before any code is written.
• Agent Manager: Manage and monitor your agents and subagents, and review their artifacts before changes land.
• Model Context Protocol (MCP) integration: Connect external databases and third-party tools by adding MCP servers from the built-in MCP Store.
• Enterprise-grade security: When you sign in with your business account, your data is protected under Google Cloud Terms of Service: it is not used to train foundation models, and you can run within your Google Cloud VPC Service Controls (VPC-SC) perimeter.

Highlights

1. Agent orchestration and the Shared Agent Harness

Antigravity coordinates background work in private, secure sandboxes, so it does not crowd your IDE or fill your prompt context.

• Dynamic parent-child subagents: For complex projects, the parent agent uses the invoke_subagent tool to spawn child agents. Each child agent starts with a clean, isolated context window for higher accuracy and lower token use.
• Flexible workspace modes: Choose how subagents work with your code:
  • inherit: Work directly in the parent's current folder.
  • branch: Spin up a dedicated, isolated Git worktree.
  • share: Run multiple subagents in a synchronized shared directory for collaborative work.
• Context-lean handoffs: Child agents complete their work, return a structured summary to the parent, and then stop. The parent's context stays small and focused.
• Specialized subagents out of the box:
  • research: Runs codebase research, file navigation, and structural exploration.
  • self: A clone of the calling agent, with the same instructions and tools.
  • Custom subagents: Define your own workflows with Markdown files and YAML frontmatter.

2. In your IDE

Antigravity works with the context already open in Visual Studio and feels like part of the IDE.

• IDE context awareness: The agent understands the files you have open, your current selection, and your solution, so its changes fit your project. Multi-root workspaces are supported.
• Native Visual Studio surfaces: Launch and track agents from Solution Explorer context menus, dockable tool windows, and a dedicated Output window.
• Interactive plans: The agent's plans and reports open in an artifact viewer with checklists you can act on and file links you can click.

3. The Agent Manager

Step back from individual lines of code to manage entire engineering workflows in the Agent Manager.

• Evidence-based artifacts: Verify the agent's work through auditable deliverables, including implementation plans and code diffs.
• Terminal command auto-execution (Agent Settings): Control how the agent runs shell commands:
  • Request Review: The agent never auto-executes commands, except those on your Allow list.
  • Always Proceed: The agent runs commands automatically, except those on your Deny list.
• Non-workspace file access: By default the agent only accesses your project folders; allow access outside the workspace only when needed.

4. Model Context Protocol (MCP) database and tool integration

Antigravity uses the open Model Context Protocol (MCP) to connect data and tools safely.

• Install MCP servers: Add servers from the built-in MCP Store, such as GitHub, Atlassian, Linear, and Notion, plus databases like BigQuery, Spanner, AlloyDB, and Cloud SQL for SQL Server.
• Configuration management: Manage servers in standard mcp_config.json files, globally (~/.gemini/config/mcp_config.json) or locally (.agents/mcp_config.json).
• Credentials stay local: Connection details live in your local mcp_config.json, and every MCP tool call is governed by Antigravity's permissions, which ask for approval before running by default.

5. Enterprise security and compliance

Antigravity runs on Google Cloud infrastructure and supports enterprise data policies.

• Data protection: With a business account, your prompts, responses, code, and telemetry run under Google Cloud Terms of Service and are never stored outside your private environments or used to train foundation models.
• VPC Service Controls (VPC-SC): Add the Agent Platform API (aiplatform.googleapis.com) to your VPC-SC perimeter to keep agent communication on private channels.
• IAM identity management: Access uses standard Google Cloud IAM. To connect, a developer needs the Agent Platform User (roles/aiplatform.user) role.
• Evidence-based auditing: Implementation plans, code diffs, and terminal command history form an auditable trail of agent actions.

Telemetry is handled in accordance with the Google Privacy Policy.