AIDEFEND in Action AIA-0106 Incident Published: Oct 26, 2026

Flowise CustomMCP: Configuration Text Becomes Server Code

Flowise builds AI workflows and connects external tools through MCP. CVE-2025-59528 lets a reachable CustomMCP configuration handler evaluate supplied text with the Node.js Function constructor. SonicWall reported exploitation in 2026. Flowise 3.0.6 replaced that evaluation with data parsing; a restricted prompt does not repair the server-side configuration boundary.

Remote Code ExecutionSupply Chain DefenseIdentity & AccessMCP SecurityAgentic AI
3 applicable AIDEFEND defenses
Source: FlowiseAI Custom MCP Node Remote Code Execution 
Author: Security News / SonicWall Capture Labs
Original article: Feb 6, 2026

Threat Analysis

  • The configuration path crosses into execution: a request supplies MCP configuration text; string processing reaches Function and grants that text the service process's JavaScript capabilities. This is application code execution, not model persuasion.
  • Access conditions matter: Internet exposure and authentication differ by deployment. The original authorized reproduction and the later exploitation report should not be collapsed into a universal unauthenticated claim.
  • The parser change is decisive: the fixed code treats the field as data. Adding an input prompt rule or merely putting the same dynamic evaluator inside a weaker sandbox leaves the incorrect trust conversion in place.

Applicable AIDEFEND Defenses (3)

AID-H-003.010
Deployed AI Software Vulnerability Remediation Lifecycle
Very High
Replace affected Flowise builds with 3.0.6 or a later supported fixed release and verify the running node-load-method handler. The decisive patch parses configuration with JSON5 instead of evaluating it with Function. In an isolated test, confirm malicious configuration text produces no process or filesystem side effect while valid MCP configuration still works; account for every deployed worker and ingress route.
AID-H-004.002
Service & API Authentication
High
For managed deployments, put Flowise administration and node-load-method routes behind a workload-OIDC gateway that verifies signature, issuer, audience, lifetime and required scope, and deny direct backend access. This blocks an unauthenticated Internet caller only where that boundary is actually enforced. The advisory reproduction uses bearer authorization, so a malicious authorized caller still requires the code-evaluation patch; do not label all affected installations universally unauthenticated.
AID-I-001.002
MicroVM & Low-Level Sandboxing
High
A Flowise operator can contain the entire vulnerable Node.js service in an externally enforced microVM or userspace-kernel runtime, with only required data mounts and no host Docker socket or management credential. Verify the actual runtime and deny a normal-container fallback. This limits host-level consequences of Function-based execution but leaves any data and credentials deliberately available inside the guest exposed. It needs a compatible managed deployment and does not replace the JSON5 parser repair.

What Defenders Should Do Now

  • Update every affected Flowise deployment and verify the effective handler, including old containers and secondary workers.
  • Restrict administration and configuration routes to authorized operators, with no public origin bypass. This reduces reachability while patch deployment is verified.
  • Use an isolated non-production fixture to test valid configuration and reject executable text. Inspect process and file effects rather than accepting an HTTP error as proof that no code ran.
  • Treat a verified microVM or userspace-kernel deployment as host containment while the parser repair is rolled out; secrets deliberately mounted inside it remain at risk.

Conclusion

AIDEFEND  connects this case to verified dependency remediation and enforced service access. The key engineering change is simple and concrete: MCP configuration remains data, and reaching its handler never authorizes arbitrary server code.

Last updated: