Docker MCP
DockerExposes the local Docker socket as tools: list containers, read logs, inspect images, start and stop services. Genuinely useful for local development,…
Exposes the local Docker socket as tools: list containers, read logs, inspect images, start and stop services. Genuinely useful for local development, and dangerous enough that the socket access deserves a deliberate decision rather than a quick install. Talks to…
Local Docker is one machine's containers; Kubernetes is a cluster with its own permission model.
Both expose container state to an agent, and the blast radius is not comparable. Docker's socket is effectively root on a single machine — acceptable on a laptop with disposable state, risky on a host that matters. Kubernetes access is mediated by RBAC and contexts, which is more governance but also more ways to point at the wrong cluster.
| Option | Install | Source |
|---|---|---|
| Docker MCP |
docker mcp gateway run
Repository ↗
|
Official |
| Kubernetes MCP |
npx -y kubernetes-mcp-server@latest
Repository ↗
|
Official |
The work is local development: start, stop, read logs, inspect a container on the machine in front of you.
The work is operating a shared cluster, and the agent needs to read pods and events with a scoped service account.
A project going archived, or a maintainer changing the install path, flips this comparison faster than any feature release. Both entries are re-checked weekly, and the date above tells you how much to trust the details.
Exposes the local Docker socket as tools: list containers, read logs, inspect images, start and stop services. Genuinely useful for local development,…
Talks to the Kubernetes API directly instead of shelling out to kubectl, so an agent can list contexts, read pod logs, and…
Claude for long documents and code; ChatGPT for the one subscription that covers images, voice, and browsing too.
Memory for facts you want to inspect; AgentDB for fuzzy recall you are willing to distrust.
Git reads the repository; GitHub runs the project. Pick by whether the question is local or social.
v0 if you already have a codebase; Replit if you do not.
Whichever tenant you already pay for. There is no meaningful winner here.
Mastra if you write Mastra; Context7 for everything else.