Radkit Remote Access

RADKit Remote Device Access

MCP Server

• Repository: CiscoDevNet/radkit-mcp-server-community
• Transport: stdio (Python via FastMCP), SSE, or HTTPS
• Requires: RADKIT_IDENTITY, RADKIT_DEFAULT_SERVICE_SERIAL, active RADKit service instance
• Python: 3.10+

How RADKit Works

RADKit provides a cloud-relayed path to on-premises devices:

NetClaw Agent  -->  RADKit Cloud  -->  RADKit Service (on-prem)  -->  Device (CLI/SNMP)

• The RADKit Service runs inside the network perimeter, onboarded with access to devices
• The RADKit Client (this MCP server) authenticates via certificate-based identity
• All communication is encrypted, relayed through Cisco's RADKit cloud infrastructure
• No direct SSH/SNMP from the agent host to the devices is needed

This is ideal for:

• Air-gapped networks where the AI agent cannot directly SSH to devices
• Cloud-hosted agents that need to reach on-premises devices
• Multi-site operations where a single RADKit service provides access to many devices
• Secure environments where certificate-based auth is required (no passwords in transit)

MCP Tools

Tool Details

get_device_inventory_names

Discovers all devices onboarded to the RADKit service. Call this first to know what devices are available.

Returns a set of device names, e.g.: {"edge-rtr-01", "core-sw-01", "dc-fw-01"}

get_device_attributes

Retrieves detailed JSON attributes for a specific device:

• Name and host address
• Device type (router, switch, firewall, etc.)
• Configuration capabilities (SSH, NETCONF, RESTCONF)
• SNMP status (enabled, community/v3 config)
• Platform details (model, OS, version)

Safe for parallel execution across multiple devices.

exec_cli_commands_in_device

Executes CLI commands on a device through the RADKit relay:

• timeout — maximum wait time per command (prevents hung sessions)
• max_lines — truncate output to N lines (prevents massive output from flooding context)
• Returns raw CLI output as text

Use this for standard show commands, debug captures, and configuration inspection.

snmp_get

Performs SNMP GET without executing CLI:

• Query one or more OIDs in a single call
• Useful for metric polling (uptime, interface counters, CPU utilization)
• Lower overhead than CLI for structured data retrieval

Common OIDs:

exec_command

Structured command execution that returns a dictionary or list:

• Includes status (success/failure) per command
• Includes truncation info if output exceeded limits
• Better for programmatic processing than raw CLI output

Workflow: Remote Device Discovery

When first connecting via RADKit:

1. Inventory: get_device_inventory_names — what devices are available?
2. Attributes: get_device_attributes for each device — type, platform, capabilities
3. Quick health: snmp_get with sysUpTime (1.3.6.1.2.1.1.3.0) for each device
4. Report: device inventory table with type, platform, and uptime

Workflow: Remote CLI Troubleshooting

When investigating an issue on a remote device:

1. Identify device: get_device_inventory_names — find the target device name
2. Check capabilities: get_device_attributes — confirm CLI access is available
3. Execute commands: exec_cli_commands_in_device with timeout and max_lines
   • show ip interface brief — interface status
   • show ip route summary — routing table health
   • show processes cpu sorted — CPU utilization
   • show logging last 50 — recent syslog messages
4. Structured output: exec_command for commands needing programmatic parsing
5. Report: troubleshooting findings with device state

Workflow: Remote SNMP Polling

When collecting metrics from remote devices:

1. Inventory: get_device_inventory_names — target devices
2. System info: snmp_get with sysDescr, sysName, sysUpTime
3. Interface status: snmp_get with ifOperStatus for key interfaces
4. Counters: snmp_get with ifInOctets, ifOutOctets for bandwidth tracking
5. Report: SNMP metric summary with uptime and interface health

Workflow: Multi-Site Health Check via RADKit

When checking health across sites served by the RADKit service:

1. Inventory: get_device_inventory_names — all devices across sites
2. Attributes: get_device_attributes for each — group by type and site
3. Health commands: exec_cli_commands_in_device per device:
   • show version — uptime, software version
   • show processes cpu | include CPU utilization
   • show memory statistics
4. SNMP baseline: snmp_get for sysUpTime, interface counters
5. Report: multi-site health dashboard with per-device status