Executive brief
IBM ContextForge MCP Gateway is a server component used to manage AI model interactions and data context. Authenticated developers can inject malicious jq filter commands that execute unrestricted operations within the server process, exposing environment variables containing database passwords, encryption keys, and other secrets that can be leveraged to take over the system as an administrator.
Technical details
The vulnerability is a jq filter injection in the jsonpath_filter field, where improper input validation allows authenticated users with developer privileges to execute unrestricted jq built-in functions like $ENV. The vulnerable component fails to sandbox or restrict jq program execution, allowing attackers to read environment variables containing JWT_SECRET_KEY, AUTH_ENCRYPTION_SECRET, DATABASE_URL, REDIS_URL, and BASIC_AUTH_PASSWORD. The attack requires prior authentication as a developer user (PR:L), involves no user interaction, and can achieve credential disclosure and privilege escalation. The fix is available in v1.0.8; deployments must upgrade and rotate all sensitive credentials listed above.
Affected products
- IBM ContextForge MCP Gateway <= v1.0.7
Timeline
- 2026-09-04: disclosed
- 2026-09-04: patched: Fix released in v1.0.8 (https://github.com/IBM/mcp-context-forge/releases/tag/v1.0.8)