PraisonAI’s MCP integration previously allowed arbitrary user-supplied commands to spawn background servers via Python’s subprocess module. The implementation forwarded the complete parent environment to the subprocess, resulting in all host process environment variables—including sensitive API keys, authentication tokens, and database credentials—being exposed to the spawned process. This exposed attackers to the potential leakage of confidential data and facilitated supply‑chain attacks by allowing untrusted code to exfiltrate secrets without additional user interaction. The vulnerability’s root cause aligns with CWE‑200 and CWE‑214.

The affected product is PraisonAI, version 4.5.127 and earlier. The MFA functionality is applicable to all deployments of the MCP protocol exposed by versions prior to 4.5.128.

The CVSS score of 5.5 indicates moderate severity, and the lack of an available EPSS score prevents precise exploitation probability assessment. PRAISONAI is not listed in the CISA Known Exploited Vulnerabilities catalog, but the nature of the flaw allows an attacker with access to the MCP command interface to retrieve environment variables, potentially compromising authentication credentials. The most likely attack vector involves an adversary controlling or influencing the command string passed to the MCP integration, which is then executed in a subprocess inheriting the host environment.