CNCF: Jaeger v2 evolves into OpenTelemetry platform for tracing AI agents with MCP, ACP, and AG-UI standards
Jaeger, a CNCF distributed tracing tool originally designed for microservices, has evolved into a platform for monitoring AI agent workloads built on OpenTelemetry. Jaeger v2 integrates three new protocols — Model Context Protocol (MCP), Agent Client Protocol (ACP), and AG-UI — and supports OpenTelemetry semantic conventions for RAG pipelines, autonomous agents, and tool call tracing.
This article was generated using artificial intelligence from primary sources.
Jaeger, the distributed tracing tool under the Cloud Native Computing Foundation (CNCF), published details of its evolution into a platform specialized for tracing generative AI applications and autonomous agents in production environments.
What is Jaeger and why does it need upgrading for AI?
Jaeger was originally designed for understanding fragmented microservice architectures — tracing requests through dozens of services and identifying bottlenecks. Modern AI agent workloads bring new complexity: RAG (Retrieval-Augmented Generation) pipelines, multi-step agents, calls to external tools, and unpredictable execution flows require specialized observability infrastructure.
Jaeger v2 has been rebuilt around the OpenTelemetry framework. Original collection mechanisms are replaced by the OpenTelemetry Collector framework, which consolidates metrics, logs, and traces into a unified deployment model and eliminates intermediate translation steps, improving ingestion performance.
What protocols does Jaeger introduce for AI agent tracing?
Jaeger v2 integrates three open-standard protocols covering different aspects of AI agent interaction:
Model Context Protocol (MCP) standardizes secure access to external data sources that AI agents use — such as databases, APIs, and documents.
Agent Client Protocol (ACP) enables uniform communication between user interfaces and AI agents, eliminating the need for custom per-agent integrations.
AG-UI (Agent–User Interaction Protocol) supports interactive workspace functionality in which engineers and AI assistants collaborate on trace analysis.
How does the ACP backend layer work for incident diagnosis?
The ACP backend layer translates natural language requests into deterministic tracing queries. An engineer can type a description like “500-level errors in the payment service with latency above 2 seconds” and the system automatically converts this into a structured query that identifies problematic spans.
The Jaeger frontend is migrating to a Zustand + React Query architecture with a built-in assistant using assistant-ui + AG-UI integration. Engineers can prompt the assistant to analyze failure paths within specific tracing spans.
Jaeger v2 also supports OpenTelemetry semantic conventions for generative AI applications: tracing RAG pipelines, autonomous agents, embedding latency, external tool calls, and token consumption monitoring. Open formats prevent vendor lock-in, which is critical for organizations wanting flexibility in infrastructure selection.
Frequently Asked Questions
- What is Jaeger and why is it important for AI agents?
- Jaeger is a CNCF distributed tracing tool that helps engineers understand complex flows across multiple services. In the context of AI agents, Jaeger v2 provides visibility into RAG pipelines, tool calls, embedding latency, and token consumption in production environments.
- What new protocols does Jaeger v2 integrate for AI?
- Jaeger v2 integrates Model Context Protocol (MCP) for secure access to external data sources, Agent Client Protocol (ACP) for uniform communication between UIs and AI agents, and AG-UI (Agent-User Interaction Protocol) for interactive workspaces.
- How does the ACP layer in Jaeger help engineers diagnose incidents?
- The Agent Client Protocol layer translates natural language requests (e.g., '500 errors in payment service with >2 second latency') into deterministic tracing queries, enabling engineers and AI agents to collaborate on failure analysis without writing complex queries manually.