Security fixes target the latest published minor release. Documented public surfaces in 1.x follow the compatibility and deprecation policy in Stability and versioning.

Do not open a public issue for a suspected vulnerability or include secrets, pairing codes, device identifiers, private logs, or unpublished exploit details in a public discussion.

Use GitHub private vulnerability reporting to share:

• the affected ADB Ready version and host environment;
• a minimal reproduction or proof of concept;
• the expected and observed behavior;
• the potential impact; and
• any known workaround.

Reports are handled privately until a fix and disclosure plan are ready. ADB Ready does not operate a bug-bounty program at this stage.

Security-sensitive areas include command execution, argument handling, target selection, file permissions, redaction, session persistence, configuration, archive contents, and release provenance. Vulnerabilities in Android Platform-Tools or a third-party runtime should also be reported to the relevant upstream project.

adb-ready mcp is a local stdio server. It does not bind a network port, make itself remotely discoverable, or upload device data. The configured AI client starts the process and decides which returned data is sent to its model provider.

The MCP surface is intentionally narrower than the CLI:

• it exposes schema-validated Android workflows, not a generic shell or raw ADB;
• one verified target is bound per connection and cannot be switched silently;
• APK paths must resolve to existing project-local files;
• UI hierarchy, logs, identifiers, and screenshots are treated as sensitive;
• mutating UI tools accept only typed actions; digest-scoped references are checked against fresh UI evidence, coordinates are display-bounded, and text uses a conservative shell-safe allowlist;
• keyboard and runtime-permission mutations require typed preconditions and fresh postconditions; unknown system or OEM dialogs fail closed;
• capture is explicit and never implied by an inspection call; and
• data clearing and app uninstall are not available as MCP tools.

MCP annotations are hints for the client approval interface, not a security boundary. ADB Ready enforces target selection, path containment, bounded output, and destructive-action policy inside the server. Users should still review the MCP configuration, pin the npm dependency, and keep mutating tools behind their client's approval or sandbox policy.

ADB itself can access USB devices, emulators, or configured network/remote ADB servers. That existing ADB authority remains the outer device-access boundary; the MCP server does not expand it.

See the public threat model for trust boundaries, protected assets, implemented controls, and residual risks.