Summary
Flowise contains a server-side request forgery vulnerability in chat-model node baseURL parameters because those values are not sufficiently validated. An authenticated user with chatflow creation or update permissions may redirect backend requests to arbitrary external, internal, or cloud metadata hosts, potentially exposing LLM provider API keys.
Affected context
The vulnerability may affect Flowise deployments that allow users with chatflows:create or chatflows:update permissions to configure chat-model nodes and their baseURL parameters. Risk increases when the Flowise server can reach internal services, cloud instance metadata endpoints, or other privileged network locations, and when provider credentials are attached to outbound model requests.
Exploit preconditions
- Attack surface: Chat-model node configuration and server-side outbound HTTP requests
- Required access: An authenticated account with permission to create or update chatflows
- Trigger point: Execution of a chatflow containing a manipulated baseURL value
- Why exploitation is possible: The application permits user-controlled request destinations without sufficiently restricting reachable hosts, schemes, or network ranges
Why this happens
The underlying issue is insufficient validation of a user-influenced URL used by a server-side HTTP client. When destination values are accepted without enforcing an allowlist, blocking private and link-local address ranges, and validating redirects, the server can be induced to make requests on behalf of the user. This behavior maps to CWE-918, Server-Side Request Forgery, because attacker-controlled input influences requests originating from a trusted backend.
Potential impact
- LLM provider API keys could be disclosed to an attacker-controlled or unintended endpoint
- Cloud instance metadata credentials may be exposed
- Internal services could receive unauthorized requests from the Flowise environment
- Network boundaries and service availability controls could be bypassed
- Sensitive responses from reachable systems could be returned to or processed by chatflows
General mitigation guidance
Mitigation generally focuses on reducing trust in user-controlled destinations
and limiting the network authority of the Flowise process.
Common defensive patterns include:
- Restricting chat-model baseURL values to approved schemes, hosts, and provider endpoints
- Blocking loopback, private, link-local, multicast, and cloud metadata address ranges after DNS resolution
- Revalidating destinations across redirects and protecting against DNS rebinding
- Separating provider credentials from user-configurable request destinations
- Applying egress filtering and network segmentation around application workloads
- Running the service with only the permissions and network access it requires
- Preventing sensitive credentials from being exposed to arbitrary outbound requests
How teams detect issues like this
Detection approaches may include:
- Static analysis for HTTP client calls that consume user-controlled URL or baseURL values
- Code review checks for missing host allowlists, private-address filtering, scheme validation, and redirect controls
- Dynamic tests covering loopback, private, link-local, metadata, alternate IP notation, and DNS-rebinding scenarios
- Monitoring outbound connections from Flowise to unexpected internal, metadata, or unapproved external destinations
- Policy enforcement that restricts application egress and limits which users can configure model endpoints
- Automated security review tools that flag server-side requests built from untrusted configuration

