Inspektor Gadget — Claude Code MCP server: features and setup
Connects Inspektor Gadget eBPF telemetry with AI coding agents to diagnose container and Kubernetes issues directly in chat.
no license updated 2026-10-29
Open on GitHubWhat Inspektor Gadget does
This server translates low-level Linux eBPF telemetry into distinct MCP tools. Functions like gadget_trace_dns and gadget_trace_tcp expose network queries, process activity, and file events enriched with pod and namespace metadata. It supports foreground inspection alongside background data capture.
AI models use these tools to isolate cluster failures. The assistant cross-references DNS lookup failures, socket resets, and memory terminations without manual log parsing.
Production clusters can run under a dedicated read-only mode to prevent unintended changes. You must supply a valid kubeconfig and have Docker available before starting the server.
What it helps with
- Pinpoint failed DNS resolutions and refused TCP handshakes inside cluster pods
- Capture kernel syscalls and file operations via eBPF probes
- Investigate container memory terminations and anomalous process spawns
- Deploy and stop low-level telemetry collectors directly from your coding assistant
Example prompts for the agent
- Check why DNS lookups are failing in the default namespace
- Trace active TCP connections for the auth-service container
- Inspect recent pod drops and show any refused socket connections
How to install Inspektor Gadget
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Inspektor Gadget
How do you install Inspektor Gadget in Claude Code?
You run the server via Docker, use a standalone binary, or deploy it straight into your Kubernetes cluster, then point your client configuration to it. A working Docker engine and an active kubeconfig are required.
What does Inspektor Gadget do?
It feeds low-level eBPF kernel telemetry into LLMs through the Model Context Protocol. The agent receives structured data covering network traffic, system calls, and container events.
Can the server be safely run in production environments?
Yes, it includes an optional read-only mode. That setting restricts agent activity strictly to non-destructive observability tools.