CVE-2026-53711
MediumCVSS 5.1Summary
MCP Context Forge versions 0.2.0 through 0.3.1 store a prompt template without sufficient output encoding before rendering in the MCP Admin UI. An authenticated attacker can save markup with script-capable event handlers that execute when another user views the saved prompt. This enables stored XSS, interface alteration, phishing, keystroke observation, and actions in the victim's browser session.
Risk Assessment
An attacker can execute script in another user's browser, hijack sessions, display fake content, or steal data. The risk affects all MCP Admin UI users.
Recommendation
Upgrade MCP Context Forge to version 0.3.1 or later. Until patched, restrict access to the Global Prompts feature and review saved prompts.
Other vulnerabilities in MCP Context Forge
See all- CVE-2026-53709Critical
MCP Context Forge is an MCP gateway and administrative platform. Prior to 1.0.2, mcpgateway/config.py and .env.example ship a known JWT_SECRET_KEY, while verify_jwt_token() accepts HS256 tokens signed with that key and get_current_user() creates a platform administrator when the token subject matches PLATFORM_ADMIN_EMAIL and REQUIRE_USER_IN_DB is false. A remote unauthenticated attacker can therefore mint accepted administrator tokens without a database account. While the vulnerable signing key remains configured and the platform-admin bootstrap remains enabled, forged access survives password rotation, account deletion, and session revocation, allowing administrative disclosure, modification, and service disruption.
- CVE-2026-53710Critical
MCP Context Forge is an AI gateway, registry, and proxy for MCP, A2A, REST, and gRPC APIs. Prior to 1.0.2, the python_sandbox_server in mcp-servers/python/python_sandbox_server/src/python_sandbox_server/server_fastmcp.py exposes raw getattr through safe_builtins, omits a required _getattr_ guard, and relies on validate_code checks for literal dangerous dunder strings. An attacker can construct dunder names at runtime, traverse the Python class hierarchy, reach subprocess.Popen, and execute OS commands with the server process privileges through the execute_code MCP tool. The HTTP/SSE transport can expose this tool without authentication, while stdio-only deployments have reduced network reachability. The issue affects the python_sandbox_server subproject and does not directly affect the core Context Forge gateway or proxy components. This issue is fixed in version 1.0.2.
Original NVD description (English source)
MCP Context Forge provides an MCP Admin UI for managing gateway resources. From 0.2.0 until 0.3.1, the Global Prompts feature stores a prompt template without sufficient output encoding before the MCP Admin UI renders it. An authenticated attacker can save markup containing script-capable event handlers, and the code executes when another user views the saved prompt. The stored cross-site scripting can alter the interface, present phishing content, observe keystrokes available to the page, and act with the viewer's browser session. This issue is fixed in version 0.3.1.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS

