AI Assistant and MCP Server

Jaeger exposes trace data to AI agents in two ways, and lets an operator teach those agents how to analyze it. Both surfaces are served by the jaeger_query extension on the query port, and both are off until the ai block is configured.

• The MCP server serves Jaeger’s telemetry tools over the Model Context Protocol at /api/ai/mcp/. Any MCP client, such as a coding agent or an IDE, can use it to search traces, map a trace’s structure, compute its critical path, and inspect individual spans. See MCP Server.
• The AI Assistant is a bring-your-own-agent framework. A gateway in jaeger_query accepts chat turns over the AG-UI protocol at /api/ai/chat, relays them to an agent of the operator’s choice over the Agent Client Protocol (ACP), and gives the agent the same MCP tools to answer with. The Jaeger UI’s chat panel is one AG-UI client of that gateway; any other AG-UI client can use it directly. See AI Assistant.
• Skills are markdown playbooks that an agent reads through the read_skill tool before it attempts a task, such as walking a failed trace to its first error. Jaeger ships built-in skills, and an operator can serve their own from a directory, so an installation can teach the assistant about its domain without rebuilding Jaeger. See Custom Skills.

Jaeger ships no agent and no model provider, and it holds no model API key. The MCP server works on its own. The assistant works only when an operator runs an agent and points Jaeger at it.

The tools are shaped for progressive disclosure: an agent first searches for candidate traces, then maps a trace without span attributes, then inspects only the spans that look suspicious. A single trace can carry thousands of spans, so this shape keeps a whole trace out of the model’s context window.

Access to either surface is access to your trace data. AI Agents to Query on the security page describes how to put authentication and TLS in front of them and how to limit the assistant to a group of users.