MCP¶

systems-manager exposes the same capability several ways: as MCP tools an agent calls, as a Python API you import, and as a CLI for direct package management. The full ecosystem role and concept map are in Overview.

As an MCP server¶

Once deployed, the server defaults to the compact intent surface and keeps consolidated action-routed modules as the governed backing capabilities. Load only the domain needed for the current request, then discover its exact action schema.

Tool	Operations
`sm_system_operations`	install / update / clean / optimize, OS + hardware statistics, package search, health check, uptime, env vars
`sm_service_operations`	manage system services
`sm_process_operations`	inspect and manage processes
`sm_network_operations`	network analysis
`sm_disk_operations`	manage disks
`sm_user_operations`	user and group management
`sm_file_operations`	file and log management
`sm_cron_operations`	manage cron jobs
`sm_firewall_operations`	firewall management
`sm_storage_health`	SMART, RAID, and optional BMC drive-fault diagnostics
`systems_ingest_host`	privacy-safe governed host graph ingestion

Agent identity, specialist registry, scheduler, and watchdog domains are also registered when their current provider toggles are enabled.

Example agent prompts that map onto these tools:

"Update the system and install git and htop" → sm_system_operations
"Show the hardware statistics for this host" → sm_system_operations
"Which processes are using the most memory?" → sm_process_operations

Tool modules are individually togglable with environment switches (for example, MISCTOOL) and can be filtered through the current Agent Utilities visibility configuration. A denied or unavailable capability must not be translated into a raw command or generic file mutation.

As a Python API¶

detect_and_create_manager() returns the appropriate SystemsManagerBase implementation for the current operating system (apt, dnf, zypper, pacman, or Windows), so the same calls work across distributions.

from systems_manager.systems_manager import detect_and_create_manager

manager = detect_and_create_manager(silent=True)

# Reads — telemetry that degrades gracefully
os_stats = manager.get_os_statistics()         # CPU, memory, load, OS metadata
hw_stats = manager.get_hardware_statistics()   # hardware inventory
logs = manager.get_system_logs(lines=100)      # recent journal / system logs

For fleet operations, delegate to an authenticated instance on the target host or compose the governed tunnel-manager capability through GraphOS. See Multi-Host.

As a CLI¶

The systems-manager console script drives package management and host maintenance directly:

# Update packages and install applications
systems-manager --update --install "git,curl,htop"

# Install Python modules
systems-manager --python "ruff,pytest"

# Clean and optimize the system
systems-manager --clean --optimize

# Display telemetry
systems-manager --os-stats
systems-manager --hw-stats

The optional helper must be provisioned by the deployment boundary; the CLI does not write sudoers policy. See Sudo Security. Run systems-manager --help for the complete flag list.