When an organization deploys AI agents at scale, high uptime and low latency are an important baseline. However, Platform owners and business stakeholders could be flying blind on several fronts:

• - **The Insights Gap:** Non-technical stakeholders have limited visibility into which AI tools are driving productivity and which are being ignored.
• - **Adoption Friction:** You might find that 80% of your developers have access to a specific tool, but only 2% are using it. Without such usage data, you can't investigate if that's due to a lack of training, a broken integration, or a tool that simply isn't useful.
• - **Unpredictable Costs:** MCP calls often involve data processing and token consumption. Without Response Body Size metrics and Method-level tracking, you cannot predict how your cloud or LLM costs will scale as you scale from 100 to 10,000 users.

Without metrics surfacing adoption trends, blocked request patterns, or tool consumption growth, platform teams can't make informed decisions about where to expand access, where to tighten policy, or how to report AI ROI to business stakeholders. Governance only kicks in after something goes wrong.

As AI systems grow more sophisticated, the connections between agents matter just as much as the agents themselves. [Model Context Protocol (MCP)](https://konghq.com/blog/learning-center/what-is-mcp)Model Context Protocol (MCP) and [agent-to-agent (A2A) communication](https://konghq.com/blog/product-releases/kong-agent-gateway)agent-to-agent (A2A) communication are becoming a core pattern in enterprise AI architecture, but a potential blind spot. With this release, Kong Konnect Analytics extends its AI visibility beyond LLM traffic to include full MCP and A2A metrics, giving platform teams a unified view.

With MCP and A2A Analytics, teams can now track request counts, response latency, upstream latency, and Kong latency — all broken down by percentile (p50, p95, p99) — and slice that data across dimensions like A2A method, task ID, context ID, agent, consumer, and status code. Whether you're investigating reports of a slow task handoff or simply understanding which agents are driving the most traffic, Konnect now gives you the data to act.

With MCP and A2A metrics alongside LLM analytics in a single place, platform teams no longer need to stitch together logs and Prometheus queries to get a picture of their multi-agent environment — that picture is now built into the gateway. As AI adoption grows across your organization, Konnect Observability grows with it: the same governance, the same telemetry pipeline, the same analytics experience your teams already rely on for the rest of your AI gateway traffic.

When a Kong customer, a large financial services firm, decided to scale Claude Code access from 1,000 to 5,000 developers, the initiative could have been hard to manage without the right infrastructure. What worked for a controlled pilot cohort would buckle under the weight of five times the users, dozens of internal data sources, and the regulatory scrutiny that follows every technology decision in financial services. 

The answer came in routing all MCP server access through Kong AI Gateway — a single, governed point that gave the platform team visibility and control.

Rather than allowing developers to connect directly to MCP endpoints in ad-hoc ways, Kong enforced authentication, applied rate limits, and logged every tool invocation, turning what could have been a sprawling shadow-AI problem into a managed, auditable system that compliance and security teams could actually get behind.

What will help make the program robust is MCP adoption metrics built on top of Kong's telemetry. By surfacing which MCP tools are called most frequently, which teams were driving the highest utilization, and where latency issues could be introducing friction, the platform team can make targeted interventions rather than guessing. High-usage tools can be prioritized for performance optimization; low-adoption teams can get dedicated support.

The metrics also gave engineering leadership the language to have honest, data-driven conversations with stakeholders. Not just saying, "We're rolling out AI." But giving a clearer picture of where it’s working and where it isn't.

Ready to see the full picture? Start building your AI governance and adoption strategy on a foundation of data.

A2A and MCP Metrics are available in the latest version of [Kong Konnect](https://cloud.konghq.com/)Kong Konnect. Try it now — or [get a demo](https://konghq.com/contact-sales)get a demo today. Or [explore the docs](https://developer.konghq.com/ai-gateway/ai-otel-metrics/)explore the docs to learn more.

*We want to hear from you. Reach out to our product team to share your feedback on how we can better support your AI journey.*