CVE-2026-55557: browse-mcp is a Playwright-based headless-browser MCP server for MCP-capable agents. Prior to 0.8.2,…
- Occurred
- Aug 25, 2026
- Published
- Aug 25, 2026
- Source
- NVD CVE
Summary
browse-mcp is a Playwright-based headless-browser MCP server for MCP-capable agents. Prior to 0.8.2, browser_download writes a fetched response body to join(save_dir, filename) without validating the caller-controlled save_dir, while browser_save_state and browser_load_state honor a caller-controlled path unchanged. A malicious MCP client, or an autonomous agent steered by indirect prompt injection on a visited page, can choose an arbitrary save_dir or state path and a URL whose response body becomes attacker-controlled file contents, allowing writes to any path the process can reach, including ~/.bashrc, autostart entries, or cron files, and potentially leading to host code execution. The force_fetch fallback also uses a raw fetch() that bypasses the BROWSE_MCP_ALLOWED_ORIGINS origin fence. This issue is fixed in version 0.8.2.
Tags
- CVE
- CVE-2026-55557
- agent-fail:injection-indirect
- agent-fail:authorization
- agent-cause:missing-guardrail
Classification
Rationale: Vulnerability in MCP agent tool allows arbitrary file writes and potential host code execution via indirect prompt injection.
Classifier: Qwen/Qwen3.5-35B-A3B-FP8