How it works
Find and read a project file
Ask your AI client to list devices, search inside an allowed project directory and read a specific source file. File access remains constrained by the local folder policy.
List my connected devices. Find README.md in the allowed project folder and summarize it.Inspect Git changes
Use read-only Git tools to inspect the current branch and diff before suggesting changes.
On my development PC, run git_status and git_diff; explain the changes without modifying files.Run and inspect tests
With terminal execution enabled and an appropriate command policy, run a project's tests and review output.
Run the project's existing tests and show failures. Do not deploy or change dependencies.Check service containers
If Docker tools are installed and permitted, inspect container state or logs. Mutations and destructive operations may require approval.
List the Docker containers and show recent errors from the API service.
Use a least-privilege workflow
Start with inspection-only tasks. Explicitly approve writes, terminals or Docker changes only when the job demands them. Do not assume your AI client itself enforces device policies.
Tool names depend on the connected integration
Representative tools include list_devices, search_files, read_file, git_status, git_diff, run_tests and docker_logs. Availability may vary with the agent version, client and permission configuration.
Frequently asked questions
Do these examples execute automatically on signup?
No. You must enroll a device, authorize a compatible MCP client and ask it to perform the task.
Can I work across multiple devices?
A client can choose a connected device for each operation. Devices do not independently delegate work to each other.