This quarter, the Amplify AI Gateway adds enterprise cost governance for AI and continues to mature Model Context Protocol (MCP) management, giving organizations the controls needed to scale AI adoption with predictable spend, stronger observability, and tighter operational oversight.
Define costs for LLM usage, including separate pricing for input and output tokens.
Set LLM quotas for applications in Fusion or Engage, based on tokens (input/output) or a monetary value (for example, "Chat Application X can spend 200 euro per month"). Enforce quotas per application, set a default currency per tenant, and publish LLM quota usage to Engage. Spread a single quota across multiple LLMs (for example, OpenAI and Anthropic) and enforce one quota across multiple APIs/MCP servers. Quota enforcement for an application is scoped to a single data plane.
A dedicated Monitor tab visualizes all MCP transactions (whether the MCP server is a Fusion server or a proxy), filterable by project, MCP server name, or tool name, with the option to hide initialization and discovery transactions.
Adds tool titles and output schemas, structured output for integrations, ResourceLinks in tool executions, and automatic protocol negotiation across supported versions (2025-11-25, 2025-06-18, 2025-03-26). Existing Fusion MCP servers can be upgraded from 2025-03-26 to 2025-11-25.
Define rate limits to control incoming requests, applied to any call against the MCP, including initialize calls and calls from exposed tools.
MCP proxy tool selection
A redesigned MCP proxy server screen lets a proxy selectively expose tools from the back-end server.
Clone MCP servers and proxies
A new Clone action quickly duplicates an existing MCP server or MCP proxy server.
Improved external MCP synchronization
When a connected external MCP server evolves, the improved synchronization presents all added, updated, and removed tools (and protocol updates) plus the proxies impacted by those changes.
OAuth Authorization Code with PKCE
MCP proxy servers can now connect to external MCP servers secured with the OAuth Authorization Code with PKCE grant type.
WIF connectivity was added to the Google Gemini connector alongside API Key, with a redesigned UI grouping IDP, Google STS, and Google IAM fields, plus a Test button to validate the configuration.