Kibana MCP Server
Access, search, and manage Kibana from any MCP-compatible client. The most widely used project we maintain.
- Data views and saved searches
- Dashboards and visualizations
- Saved objects
- Detection rules and alerts
- Cases and timelines
- Security and observability queries
- Full Kibana API access
- Elastic Security investigation
- Dashboard analysis
- Kibana object management
- AI agents operating Kibana
- Internal copilots and automation
Kibana exposes a large HTTP API, and almost none of it is reachable from an AI client without writing a bespoke integration first. This server closes that gap: it presents the Kibana API to any MCP-compatible client — Claude Desktop, an internal agent, a CI job — so an assistant can read a data view, pull a detection rule, open a case, or fetch the objects behind a dashboard without a human copying JSON between windows.
The tool surface is generated from Elastic’s own OpenAPI specification for the Kibana API (Elastic Stack 8.x), so it tracks the documented API rather than a hand-curated subset that drifts out of date.
What it is useful for
The pattern that comes up most often is investigation. An analyst asks a question in natural language; the agent resolves it into the actual Kibana calls — fetch the alert, fetch the rule that produced it, fetch the timeline, fetch the related cases — and returns the assembled context. The analyst still makes the call. The mechanical part of pulling evidence together stops being manual.
The second pattern is object management. Kibana estates accumulate saved objects: data views nobody owns, dashboards that reference deleted indices, rules duplicated across spaces. Being able to enumerate and cross-reference those through an agent turns a half-day audit into a conversation.
Installation
# Global installation
npm install -g @tocharianou/mcp-server-kibana
# Or run directly
npx @tocharianou/mcp-server-kibanaConfiguration for Claude Desktop and other MCP clients, along with the full tool list and authentication options, is in the repository README.
Deploying it inside an enterprise
The published package is deliberately unopinionated about authorization — it uses the credentials you give it, and inherits whatever that Kibana user can do. That is the right default for a local workstation and the wrong one for a shared production estate.
When we deploy this for customers the work is mostly the part that is not in the repository: mapping the agent to a dedicated Kibana role instead of a human’s account, splitting read tools from write tools, putting approval gates in front of anything that mutates a rule or closes a case, scoping data access to the spaces and indices that engagement is allowed to touch, and logging every tool call somewhere the security team can review it.
Community-maintained. Not an official product of Elastic or of the MCP project.