Security vulnerabilities and automated fixes for mcp issues
4 posts found
A critical input validation vulnerability was discovered in `src/mcp/presets/commerce/inputs.ts` where the `normalizeCommerceAccountInput` function accepted loosely-typed `Record<string, any>` arguments without verifying field types or object structure. This allowed attackers to inject malicious payloads through MCP tool invocations. The fix adds explicit type guards and structural validation to ensure only properly-typed string values reach downstream consumers.
The `@modelcontextprotocol/sdk` package (version 0.5.0) shipped without DNS rebinding protection enabled by default, leaving any Node.js application that hosts an MCP server exposed to cross-origin attacks from malicious websites. Upgrading to version 1.24.0 enables host-header validation and brings a suite of new security dependencies—including `cors`, `express-rate-limit`, and `jose`—that collectively close the attack surface. This fix was identified and patched automatically by Orbis AppSec b
The Model Context Protocol (MCP) TypeScript SDK in version 1.17.1 lacked DNS rebinding protection, leaving applications vulnerable to a sophisticated network-level attack. By upgrading to version 1.24.0, DNS rebinding protection is now enabled by default, preventing attackers from exploiting the SDK's HTTP request handling to access internal resources or bypass security controls.
A high-severity vulnerability (CVE-2026-27896) in the Model Context Protocol Go SDK v1.3.0 allowed attackers to bypass security controls through improper handling of case sensitivity. The fix upgrades the dependency from v1.3.0 to v1.3.1, which correctly normalizes case comparisons. This vulnerability was particularly concerning for CLI tools where attackers could manipulate input to evade validation logic.