NewApp Monitoring is liveSign up free
AgentbeamBeta
AI agent vulnerability · Other MCP servers and tools

CVE-2026-61459: mcp-server-kubernetes argument injection can expose Kubernetes cluster credentials

criticalCVSS 4.0 9.3 · 3.1 9.8GHSA-wmg3-h8mf-wgvr

mcp-server-kubernetes before 3.9.0 lets an attacker slip a `--server` flag into the kubectl_get, kubectl_describe and kubectl_delete tools, which sends kubectl to an attacker-run API server and leaks the operator's bearer token. That can lead to full cluster compromise. Upgrade to 3.9.0.

Published Jul 10, 2026 · Updated Oct 28, 2026 · Analysis Nov 1, 2026

What an attacker can do

The structured tools kubectl_get, kubectl_describe and kubectl_delete build a kubectl command line from the resourceType and name parameters. The server has an assertNoDangerousFlags check meant to block unsafe options. Values that begin with a dash get past it, so kubectl treats them as flags instead of resource names. This is argument injection (CWE-88).

The useful flag for an attacker is --server. It points kubectl at a different API server. kubectl then sends the request, with the credentials from the operator's kubeconfig, including the bearer token, to the host the attacker named. With that token the attacker can act against the real cluster. VulnCheck rates the result as full cluster compromise, and the CVSS vectors need no privileges or user interaction.

The attacker needs to influence the tool arguments. In an MCP setup that means anything that steers the AI agent calling the tool: a prompt injection in a ticket, a README, a log line or a web page, or a malicious MCP client. The agent does not have to be tricked into running a visibly dangerous command, because the arguments look like ordinary resource names.

Who is affected

The npm package mcp-server-kubernetes in every version below 3.9.0. Servers wired into an AI client with a kubeconfig that holds a real token are the exposed setup. Check what you run:

npm ls -g mcp-server-kubernetes
npm ls mcp-server-kubernetes

Also look at your MCP client configuration for a pinned version or an npx call without one.

mcp-server-kubernetes

Ecosystem
npm
Affected
< 3.9.0
Fixed in
3.9.0
Check
npm ls mcp-server-kubernetes

How to fix it

Upgrade to 3.9.0:

npm install -g [email protected]

The release page for 3.9.0 carries no notes, and the sources give no workaround. Because the token is exposed when the flaw is triggered, rotate the kubeconfig credentials used by any older install that agents could reach. Prefer a read-only, namespace-scoped service account for the agent.

How to tell if you were hit

Review MCP tool-call history for kubectl_get, kubectl_describe or kubectl_delete calls where resourceType or name starts with -, especially --server, --server= or -s. Check egress logs and DNS for connections from the host running the server to unknown Kubernetes API endpoints, and check the real API server's audit log for token use from unexpected addresses. Beam records the MCP tool calls an agent makes, so calls with a leading-dash argument can be searched for there.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

MCP Server Kubernetes before 3.9.0 contains an argument injection vulnerability in structured tools (kubectl_get, kubectl_describe, kubectl_delete) that allows attackers to bypass the assertNoDangerousFlags security check by supplying resourceType and name parameters with leading dashes. Attackers can inject the --server flag to redirect kubectl commands to an attacker-controlled API server, causing the operator's bearer token to be transmitted externally and enabling full cluster compromise.

Credited to PercevalFox, devjoinedthechat, wlassalle724.

Questions

Does the attacker need cluster credentials first?

No. The attacker only needs to control the tool arguments. The flaw makes kubectl send the operator's own token to a server the attacker controls.

Which tools are affected?

The structured tools kubectl_get, kubectl_describe and kubectl_delete, through their resourceType and name parameters.

Is CVE-2026-61459 exploited in the wild?

No source says so. VulnCheck links a public disclosure issue, and the EPSS score is about 2.4%.

Is a fixed version available?

Yes, 3.9.0. Its release page carries no notes, so check the linked commit if you need details.

References

How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology

More Other MCP servers and tools vulnerabilities

All Other MCP servers and tools entries