Google Antigravity Python SDK — v0.1.1 Release Notes

[allenhutchison]

Google Antigravity Python SDK — v0.1.1 Release Notes

This release focuses on significant enhancements to the Model Context Protocol (MCP) integration, adds native Vertex AI authentication support, improves robustness with better error handling, and includes critical fixes for token usage tracking.

🌟 Key Highlights

• Enhanced MCP Integration: Major updates to MCP server configurations, including automatic namespaced tool prefixing to prevent collisions, tool filtering (allowlist/denylist), and overloaded safety policies for easier permission management.
• Vertex AI Authentication: Integrated native support for Vertex AI authentication in the Python SDK.
• Bug Fix for Usage Tracking: Resolved an issue where token usage metadata was not correctly populated when structured output was requested.

---

📋 Detailed Changes

Model Context Protocol (MCP)

• Tool Prefixing: The SDK now automatically namespaces and prefixes MCP tools (using the pattern mcp_{server_name}_{tool_name}) to prevent name collisions when connecting multiple MCP servers.
• Tool Filtering: Added support for enabled_tools (allowlist) and disabled_tools (denylist) in server configurations to limit which tools are exposed to the agent.
• Simplified Safety Policies: Overloaded safety policy helpers (policy.allow, policy.deny, and policy.ask_user) to accept the MCP server configuration object directly. This simplifies defining permissions for all or specific tools on a server.
• Configuration Validation: The name field is now mandatory in MCP server configurations, and tool/parameter names are validated to ensure they are valid Python identifiers.

Features & Enhancements

• Vertex AI Auth: Integrated native support for Vertex AI authentication in the Python SDK.

Robustness & Usability

• Improved Error Handling: The SDK now raises explicit, descriptive exceptions for terminal errors rather than failing silently.
• Structured Output Fix: Fixed a bug where token usage_metadata was not correctly returned when structured output (response.structured_output()) was used.

Bug Fixes

• Fixed type-checking (pytype) warnings (including wrong-keyword-args in LocalAgentConfig) across several modules.