Malicious MCP Server Configuration in PRs Enables Remote Code Execution and Secret Exfiltration

Medium5.3CVE-2026-47751 · Published May 20, 2026

Due to the combination of checking out PR head branches (attacker-controlled), reading `.mcp.json` from the working directory via default setting sources, and unconditionally enabling all project MCP servers via `enableAllProjectMcpServers`, it was possible for an attacker who opened a PR containing a malicious `.mcp.json` file to achieve arbitrary code execution on the GitHub Actions runner. This could lead to exfiltration of secrets available to the workflow (such as API keys and tokens) when a privileged user triggered the Claude action on the PR. Exploiting this required the ability to open a pull request against a repository using the claude-code-action and a privileged user or automatic trigger to invoke the action on that PR. Users pinned to a vulnerable version of claude-code-action are advised to update to the latest version. Users referencing anthropics/claude-code-action@v1, anthropics/claude-code-action@beta, anthropics/claude-code-action@main, or other non-pinned tags will have received this fix already Thank you to hackerone.com/reptou for reporting this issue.