CVE-2026-130007: k8s-mcp-server let anyone on the network run Kubernetes tools without a login

The k8s-mcp-server Model Context Protocol server exposed Kubernetes management tools over HTTP without authentication, allowing any network client to execute cluster operations with the server's privileges.

Christopher Ort
By Christopher Ort · 3 min read
Share
A security researcher at a laptop reviewing network traffic logs showing unauthorized API calls to a Kubernetes cluster management interface
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. What you should do
  5. What it means
  6. Sources
i10X Quick Take

Quick Take

  1. What: k8s-mcp-server versions before 1.1.0 accepted unauthenticated MCP tool calls over SSE and streamable HTTP transports
  2. Who is affected: Deployments running k8s-mcp-server in SSE or streamable-HTTP mode (the default was SSE on 0.0.0.0:8080)
  3. What to do: Upgrade to version 1.1.0 or later, which requires bearer authentication and binds to loopback by default
  4. Severity: High (CVSS 7.9), any network client could enumerate and invoke Kubernetes tools with the server’s cluster privileges
  5. Status: Fixed in version 1.1.0 released August 31, 2026

In plain English

k8s-mcp-server is a tool that lets AI agents manage Kubernetes clusters through the Model Context Protocol. Before version 1.1.0, when the server ran in its default SSE mode, it listened on all network interfaces and did not check who was connecting. Anyone who could reach the port could list and run Kubernetes commands (including destructive operations like deleting resources or installing Helm charts) with the same permissions the server had. Version 1.1.0 fixes this by requiring a secret token for every request and only listening on the local machine by default.

What you should do: If you use k8s-mcp-server, upgrade to 1.1.0. If you do not use this tool, you can ignore this issue.

What happened

k8s-mcp-server is an open-source Model Context Protocol server that exposes Kubernetes management tools to MCP clients. Prior to version 1.1.0, the server defaulted to SSE (Server-Sent Events) mode, listening on 0.0.0.0:8080 with no authentication middleware. The GET /sse handler created a sessionId for any caller, but that sessionId was only a correlation value, not a credential. The POST /message handler accepted any loadable sessionId and dispatched tools/call requests without checking an Authorization header, bearer token, or Origin.

Write-capable Kubernetes tools (Create, Delete, RolloutRestart) and Helm tools (Install, Upgrade, Uninstall, Rollback) were all registered by default unless the server was started with the -read-only flag. An unauthenticated attacker with network access to the port could enumerate the full tool list and invoke any registered Kubernetes operation with the server’s configured cluster privileges.

Security researcher bebold6133 opened GitHub issue #131 on August 13, 2026, asking for a private channel to report the flaw, and the maintainer then enabled private vulnerability reporting. The fix was merged in pull request #134 on August 19, 2026, and released as version 1.1.0 on August 31, 2026.

What you should do

Upgrade to k8s-mcp-server version 1.1.0 or later. The fixed version requires a minimum 32-character MCP_AUTH_TOKEN environment variable for SSE and streamable-HTTP transports, enforces bearer authentication before every MCP HTTP endpoint, validates browser origins through an MCP_ALLOWED_ORIGINS allowlist, binds HTTP transports to 127.0.0.1 by default, and changes the default transport to stdio. The unauthenticated /healthz endpoint remains available for container health checks but does not expose MCP functionality.

If you cannot upgrade immediately, do not expose the SSE or HTTP port to untrusted networks. Version 1.1.0 itself makes stdio the default transport, and write-capable tools stay registered unless the server is started with the -read-only flag, so running in stdio mode with -read-only reduces what an exposed server can do.

What it means

This vulnerability is tracked as CVE-2026-130007 and classified as CWE-306 (Missing Authentication for Critical Function). The GitHub security advisory is GHSA-m932-63jc-4jqq. The issue affected all deployments running k8s-mcp-server in SSE or streamable-HTTP mode before version 1.1.0, particularly containerized deployments where the default 0.0.0.0 bind exposed the port to the network.

The fix was merged through pull request #134 as commit 817d810. According to the advisory, any client able to reach the port could invoke the registered Kubernetes tools with the server’s configured cluster privileges, and the NVD entry rates the issue CVSS 7.9.

Unauthenticated HTTP modes keep exposing MCP servers that hold real infrastructure access: see our reports on Dynatrace MCP Server allowing unauthenticated tool calls in HTTP mode and Kubeshark MCP server flaws that let AI agents hijack cluster images. New cases are tracked in our AI Security hub.

Sources

  • NVD CVE-2026-130007
  • GitHub Security Advisory GHSA-m932-63jc-4jqq
  • GitHub Issue #131
  • GitHub Pull Request #134
  • Fix commit 817d810
  • Release v1.1.0
Christopher Ort

Christopher Ort

Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU). He is the author of “The Style Penalty”, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.

Editor's picks

  1. GAO Report: ChatGPT, Claude, Gemini and Grok Produced Viable Attack Plans After Safeguards Were Bypassed
  2. ResolveAI SRE-1: A 27B Model Trained to Find the Root Cause of Production Incidents
  3. Reson8 Rumble-2-Flash: Open-Weight 0.5B Speech Recognition Model Covers 40+ Languages
  4. Pika Enact Lets Creators Direct AI Video Actors and Cameras in Real Time
  5. Google Stitch Loop: An Always-On Product Agent That Fixes Security, Speed and Accessibility via Pull Requests