Multiple security vulnerabilities have been identified in Kubeshark, affecting the Model Context Protocol server functionality in releases from v53.1.0 through v53.3.0. The issues involve unsafe handling of caller-controlled tool arguments and can allow an attacker who influences agent-supplied input to alter deployment behavior or write data to unintended local locations. The overall risk is high because the flaws can lead to privileged container compromise, cluster-admin access, or local file integrity loss.
CVE-2026-129236 (CVSS 7.2 — High): The MCP start_kubeshark tool appends a caller-supplied pod regex argument without an end-of-options separator, allowing a malicious value to be interpreted as command-line flags. An attacker who can influence agent-driven tool arguments can inject configuration overrides that redirect deployed images to an attacker-controlled registry, potentially leading to privileged container execution and cluster-admin compromise.[/subscribe_to_unlock_form]
Multiple security vulnerabilities have been identified in Kubeshark, affecting the Model Context Protocol server functionality in releases from v53.1.0 through v53.3.0. The issues involve unsafe handling of caller-controlled tool arguments and can allow an attacker who influences agent-supplied input to alter deployment behavior or write data to unintended local locations. The overall risk is high because the flaws can lead to privileged container compromise, cluster-admin access, or local file integrity loss.
CVE-2026-129236 (CVSS 7.2 — High): The MCP start_kubeshark tool appends a caller-supplied pod regex argument without an end-of-options separator, allowing a malicious value to be interpreted as command-line flags. An attacker who can influence agent-driven tool arguments can inject configuration overrides that redirect deployed images to an attacker-controlled registry, potentially leading to privileged container execution and cluster-admin compromise.[emaillocker id="1283"]
CVE-2026-129235 (CVSS 5.3 — Medium): The MCP download_file tool writes fetched data to a caller-supplied destination without validating the path, allowing an attacker who can influence tool arguments to write bytes to arbitrary local paths writable by the process. The same tool can also request arbitrary Hub API endpoints, enabling access to captured traffic or other sensitive data on deployments where Hub authentication is not enforced.
These vulnerabilities collectively present a significant risk to environments that expose the affected Model Context Protocol server to untrusted agent content or indirect prompt injection. The most exposed users are operators running the server with elevated deployment permissions or with local credentials that can access sensitive cluster and network data. They should prioritize review of affected deployments and limit the server's exposure to untrusted inputs.
We recommend you to update Kubeshark to version 53.4.0 to address CVE-2026-129236 and CVE-2026-129235.
The following reports contain further technical details:
[/emaillocker]