A path traversal vulnerability exists in the prepare_kaggle_dataset function of dexhunter/kaggle-mcp. The function accepts a competition_id parameter without adequate validation, allowing an attacker to include directory traversal components that resolve to locations outside the intended dataset directory. If exploited, the attacker can read arbitrary files on the filesystem where the service runs, potentially exposing sensitive data. The flaw corresponds to CWE‑22.

The vulnerability affects all releases of dexhunter/kaggle-mcp up to the code revision 406127ffcb2b91b8c10e20e6c2ca787fbc1dc92d. Because the project uses a rolling release strategy, no specific version numbers are available; any current revision that has not been updated by the maintainer remains vulnerable.

The vulnerability is exploitable remotely via the public API endpoint that processes competition_id inputs. The documented CVSS score is 6.9, indicating medium severity with potential confidentiality and integrity impact. No EPSS score is available and the issue is not listed in CISA KEV, but the fact that it is publicly disclosed and the project has not yet released a fix suggests that the opportunity for exploitation may remain until the maintainer publishes a patch. Security teams should treat the risk as moderate to high for exposed services.