How jerrelblankenship/jb-kibana-mcp works
Kibana MCP Server exposes saved Kibana content and Elasticsearch search through MCP. AI clients connect over either stdio or HTTP/SSE; the transport hands MCP requests to shared server logic, which registers read-only resources and executable tools. Those handlers invoke the Kibana client for dashboards, visualizations, data views, saved searches, dashboard export, and log queries. The client authenticates REST requests to Kibana, which also proxies searches to Elasticsearch. README documents capabilities beyond the sampled handler excerpts; those are represented at the subsystem level where wiring is not fully evidenced.
Main entry points: http-server.ts HTTP/SSE Entry · server.ts MCP Server · client.ts Kibana API Client · index.ts stdio Entry
Components 10
MCP Transports
- stdio Entry
src/index.ts - HTTP/SSE Entry
src/http-server.ts
MCP Capabilities
- MCP Server
src/server.ts - Read Resources
src/resources/index.ts - Executable Tools
src/tools/index.ts
Kibana Integration
- Kibana API Client
src/kibana/client.ts - Authentication
src/kibana/auth.ts
Other components
- AI Assistant
- Kibana REST API
- Elasticsearch
Connections 13
- AI Assistant → stdio Entry: starts process
- AI Assistant → HTTP/SSE Entry: connects MCP
- stdio Entry → MCP Server: creates and connects
- HTTP/SSE Entry → MCP Server: creates per session
- HTTP/SSE Entry → AI Assistant: streams MCP
- AI Assistant → HTTP/SSE Entry: posts messages
- MCP Server → Read Resources: registers handlers
- MCP Server → Executable Tools: registers handlers
- Read Resources → Kibana API Client: reads Kibana data
- Executable Tools → Kibana API Client: executes API calls
- Kibana API Client → Authentication: builds request headers
- Kibana API Client → Kibana REST API: REST requests
- Kibana REST API → Elasticsearch: proxies searches