Agentic Observability With NGINX: Open Source MCP Traffic Monitoring

Key Concepts

• MCP (Model Context Protocol): An open standard developed by Anthropic that allows AI agents to communicate with LLMs and external tools/resources via a common interface.
• NJS (NGINX JavaScript): A scripting language module for NGINX that allows for the extension of NGINX functionality using JavaScript, which compiles into high-performance C code.
• Agentic Observability: The practice of monitoring AI agents to track their tool usage, latency, error rates, and traffic flow.
• OTel (OpenTelemetry): A framework for collecting and exporting telemetry data (metrics, logs, traces) to observability backends.
• NGINX Gateway Fabric: An NGINX-based implementation of the Kubernetes Gateway API.

---

1. Main Topics and Key Points

The video introduces a new Agentic Observability module for NGINX, designed to provide visibility into AI agent traffic. As AI agents increasingly use the Model Context Protocol (MCP) to interact with tools, organizations face a "blind spot" regarding how these agents behave.

• Functionality: NGINX acts as a proxy between the MCP client and the MCP server. By intercepting these calls, the module captures metadata such as tool names, error codes, and latency.
• Technical Implementation: The module leverages NJS combined with native NGINX configuration language. Because NJS compiles to C, the module maintains high performance without sacrificing speed.
• Data Export: The module utilizes NGINX’s built-in OTel exporter to ship captured telemetry data to external systems like Prometheus or Syslog servers.

2. Real-World Applications and Demos

• Observability Stack: The F5 team provides a repository that includes a full demo walkthrough. This demo spins up an NGINX instance with the NJS module, exports metrics to Prometheus, and visualizes the data using a pre-built Grafana dashboard.
• Visibility: The primary use case is solving the lack of transparency in AI agent workflows, allowing developers to audit which tools an agent is invoking and identify performance bottlenecks.

3. Methodologies and Frameworks

• Proxy Pattern: NGINX is positioned as an intermediary layer. By sitting between the MCP client and server, it inspects the traffic flow without disrupting the agent's operation.
• Extensibility: The solution is designed to be modular. Users can update their existing NGINX Open Source or NGINX Plus environments to include this functionality.
• Future Roadmap: While currently available for standard NGINX deployments, there is a long-term strategy to integrate this observability into NGINX Gateway Fabric to support Kubernetes-native AI workloads.

4. Key Arguments and Perspectives

• Addressing the "Blind Spot": Jason Williams emphasizes that the rapid rise of AI agents has outpaced current monitoring capabilities. The module is presented as a necessary tool for enterprise-grade AI, where knowing "where agents are going and what tools they are using" is critical for security and performance.
• Performance Priority: A significant argument made is that the use of NJS ensures that adding observability does not degrade the performance of the NGINX data plane.

5. Notable Quotes

• "We have agents that use MCP to be able to reach out through tools or function calling to be able to access resources they need... by NGINX actually sitting between an MCP client and MCP server, we can actually capture and observe those agents." — Jason Williams
• "Since NGINX JavaScript is basically fully... compiled into actual C that runs on NGINX, this module should still be very performant." — Alexandra Falgaria

6. Synthesis and Conclusion

The introduction of the MCP observability module for NGINX represents a strategic move by F5 to support the growing ecosystem of AI agents. By providing an open-source, high-performance way to monitor MCP traffic, the team is enabling developers to gain critical insights into AI tool usage and system health.

Actionable Takeaways:

• Availability: The module is open-source and available for both NGINX Open Source and NGINX Plus.
• Resources: Users can find the code and documentation in the nginx-mcp repository on GitHub.
• Community Engagement: Feedback and contributions are encouraged via the NGINX community forum (community.nginx.org) to help evolve the module as the AI landscape changes.