Exploring observability with MCP

Transcript

1. Exploring Observability with MCP Tiffany Jernigan tiffanyfay.dev Senior Developer Advocate Grafana Labs
2. What is Observability? Observability is the process of making a systemʼs internal state more transparent.
3. 3 4 Pillars of Observability • Metrics • Logs • Traces • Profiles
4. LLM 1.0: Access internet context using language LLM In Out Context Instructions
5. Single call, single action, chat response LLM Explanation Flamegraph
6. But LLMs only know knew what they’ve seen in training They can’t couldn’t interact with Grafana They don’t didn’t know about your dashboards They can’t couldn’t query your metrics, logs, ...
7. Making LLMs speak Grafana
8. Tool calling LLMs to go beyond chat response LLM In Out Context Instructions Tools Call Response Private context + Custom actions
9. Model Context Protocol: USB-C for AI applications https://norahsakal.com/blog/mcp-vs-api-model-context-protocol-explained/
10. Query data sources (Prometheus, Loki) Fetch alerts that are firing Find out who’s on call (IRM) Search & update dashboards Create and manage incidents Run and view Sift investigations Interact with the Asserts graph and much more to come Grafana MCP Server exposes core tools to LLMs
11. Let LLMs answer questions with Grafana context
12. Demo And thanks to Ben Sully, Ioanna Armouti, Rob Whelan, Bryan Huhta, Joe Elliott