INCIDENT MAP

Browse / Indirect Prompt Injection / CVE-2025-54994

@akoskm/create-mcp-server-stdio is vulnerable to MCP Server Command Injection through `exec` API

CVE-2025-54994 ยท 2025-09-08 critical

Class
Indirect Prompt Injection (filed by a model that read the advisory)
Technique
Code and command execution
Also involves
MCP and Tool Server Flaws
Affected
@akoskm/create-mcp-server-stdio (npm)
CVSS
CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N
EPSS
1.3% chance of exploitation in 30 days

How it works, in short

Prompt-injected input to an MCP tool is interpolated into a shell exec call, so shell metacharacters run extra commands on the server. (Our summary, written by a model. The source text is below.)

Details

Command Injection in MCP Server

The MCP Server at https://github.com/akoskm/create-mcp-server-stdio is written in a way that is vulnerable to command injection vulnerability attacks as part of some of its MCP Server tool definition and implementation.

Vulnerable tool

The MCP Server exposes the tool which-app-on-port which relies on Node.js child process API exec which is an unsafe and vulnerable API if concatenated with untrusted user input.

Vulnerable line of code: https://github.com/akoskm/create-mcp-server-stdio/blob/main/src/index.ts#L24-L40

server.tool("which-app-on-port", { port: z.number() }, async ({ port }) => {
  const result = await new Promise<ProcessInfo>((resolve, reject) => {
    exec(`lsof -t -i tcp:${port}`, (error, pidStdout) => {
      if (error) {
        reject(error);
        return;
      }
      const pid = pidStdout.trim();
      exec(`ps -p ${pid} -o comm=`, (error, stdout) => {
        if (error) {
          reject(error);
          return;
        }
        resolve({ command: stdout.trim(), pid });
      });
    });
  });

Exploitation

When LLMs are tricked through prompt injection (and other techniques and attack vectors) to call the tool with input that uses special shell characters such as ; rm -rf /tmp;# (be careful actually executing this payload) and other payload variations, the full command-line text will be interepted by the shell and result in other commands except of ps executing on the host running the MCP Server.

Reference example from prior security research on this topic:

!Cursor defined MCP Server vulnerable to command injection

Impact

User initiated and remote command injection on a running MCP Server.

Recommendation

References and Prior work

##

Disclosed by Liran Tal

References

Source: GitHub Advisory Database, used under CC BY 4.0.