Summary
Microsoft UFO exposed Streamable HTTP MCP services for mobile automation without authentication, allowing unauthenticated remote callers to interact with an ADB-connected Android device. Depending on network exposure and device permissions, an attacker could view screen or UI data, invoke device actions, and alter application or device state.
Affected context
The issue occurs in UFO deployments that enable the mobile data collection or mobile action MCP servers and expose their HTTP listeners on reachable TCP services. Risk is greatest when the listeners are accessible beyond a tightly controlled local network and when an ADB-connected Android device is available to the service.
Exploit preconditions
- Attack surface: Streamable HTTP MCP services exposed by the mobile automation server
- Required access: Network reachability to the exposed service; no application authentication is required
- Trigger point: Invocation of mobile data collection or device-action MCP operations
- Why exploitation is possible: The services accept remote requests without verifying caller identity or authorization before forwarding operations to the connected Android device
Why this happens
The underlying flaw is the deployment of network-accessible control endpoints without an authentication and authorization boundary. This maps primarily to CWE-306: Missing Authentication for Critical Function, because sensitive device operations can be invoked without establishing the caller's identity. It also relates to CWE-862: Missing Authorization, since requests are not evaluated against permissions for specific data-access or device-control actions.
Potential impact
- An unauthenticated attacker could capture screenshots or retrieve UI-tree information
- Sensitive screen, application, or device data could be disclosed
- Device input and navigation state could be changed
- Applications could be launched or interacted with without user approval
- Automated workflows could be disrupted or manipulated
- The connected Android device could be used as a stepping stone for broader compromise, depending on its permissions and network placement
General mitigation guidance
Mitigation generally focuses on placing a strong identity and authorization boundary in front of all mobile MCP services.
Common defensive patterns include:
- Require mutually authenticated or otherwise strongly authenticated connections
- Enforce per-operation authorization for screenshots, UI inspection, input, application launch, and key events
- Bind services to trusted interfaces and restrict exposure with network policies or firewalls
- Prefer secure-by-default configurations that do not publish control endpoints broadly
- Reduce trust in network location by treating every request as untrusted until authenticated and authorized
- Run automation services and connected devices with the minimum permissions required
- Log and rate-limit sensitive operations, with explicit approval controls for high-impact actions
How teams detect issues like this
Organizations can look for:
- Static analysis and code review findings where HTTP or MCP handlers invoke device-control functions without authentication middleware
- Server startup configurations that bind mobile control services to broadly reachable interfaces
- Missing authorization checks before screenshot, UI inspection, input, launch, or key-event operations
- Network scans and service inventories identifying exposed mobile MCP listeners
- Access logs containing unauthenticated requests or unusual sequences of device-control operations
- Policy checks requiring authentication, authorization, and restricted network placement for automation endpoints
- Automated security review tools that flag unauthenticated sensitive HTTP routes and missing access-control enforcement

