T H E  I D E A

Your Team Shouldn’t Have to Know How OOB Works to Use It

Out-of-band access has always demanded more from users than it should. Learn the portal. Find the device. Know which port it’s on. Memorize the CLI. That overhead slows down every engineer who isn’t the OOB specialist, and in an incident, every minute counts.

Percepxion with MCP removes that friction. Your team connects the AI assistant they already use, Claude, Copilot, or any MCP-compatible client, and from that point on, getting into a device or pulling information about it is just asking for it. The OOB layer is there. It works. They don’t have to think about it.

“Set it up once. Now everyone on the team can reach any device, in their own words.”

The Percepxion MCP Server is Lantronix’s open-source implementation of the Model Context Protocol, the emerging industry standard for connecting AI assistants to real-world systems.

H O W  I T  W O R K S

A Secure Bridge Between Your AI and Your Fleet

The MCP server runs as a local process alongside your AI client. When the AI needs device data or wants to trigger an operation, it calls a named tool on the MCP server, which validates the request, translates it into a Percepxion API call, and returns the result.


The MCP server enforces read-only mode by default, write commands require an explicit environment flag. All CLI commands dispatched to devices are logged to stderr with device ID and command string for audit purposes. Your credentials never touch the AI client.

T H E  P R O B L E M  I T  S O L V E S

Getting Into a Device Shouldn’t Require an Expert

When a device needs attention, the wrong question is “who knows how to navigate the OOB portal?” Every engineer, every NOC analyst, every operations team member should be able to get in, or get answers, without a training course and without a specialist on the phone.

The Device They Need. The Answer They’re Looking For. In Their Own Words.

Example prompt

"Get me into the Chicago datacenter core switch. I need to check the routing table."

Example prompt

"The Austin branch firewall went unreachable during last night's maintenance. What changed and when did it drop off?"

Example prompt

"The lab server in rack 3 is unresponsive. Can you reboot it? I don't have direct access."

They Ask About a Device. They Get an Answer. They Never Touch the OOB System.

A NOC analyst gets an alert. A router at a branch site is unreachable. In the old model, they’d escalate to whoever knows the out-of-band system, find the device in the portal, connect to its console, figure out what’s wrong. That takes time nobody has during an incident.

With Percepxion MCP, they open their AI assistant and ask. The AI connects through the OOB layer, pulls the audit log, checks the interface status, and tells them what changed, in plain English, in seconds. The OOB infrastructure did its job. They never had to know it was there.

• Works for any team member, not just the OOB specialist
• No portal login, no device lookup, no syntax to remember
• Full audit trail of every action taken through the AI
• The OOB admin controls what each user can do

F O R  T H E  O O B  A D M I N

You Control What Happens. Your Team Gets the Access They Need.

The abstraction doesn’t mean loss of control. The OOB admin configures the MCP server once, setting access policy, defining which operations require write permissions, blocking destructive commands, and connecting credential storage to the team’s existing secret manager.

From that point, every team member benefits. And every action they take through the AI is logged, auditable, and bounded by the policy you set.

• Read-only by default, diagnostic use needs no configuration
• Built-in block list: reload, factory-reset, write erase always blocked
• Configurable permit list to restrict exactly which commands pass
• Every device command logged for audit with device ID and timestamp

Your Team’s Secret Store. Not a Plaintext File.

For individual use, credentials in a local .env file are fine. For shared deployments or CI pipelines, the MCP server supports HashiCorp Vault and AWS Secrets Manager as credential backends, no credentials on disk, no rotation coordination.

Switch providers at runtime with reconfigure_credentials without restarting the server. The active provider is tracked separately from the startup config.

• Three providers:env(default),vault,aws
• Vault: KV v2 path configurable via env var
• AWS: reads from Secrets Manager by secret name
• Runtime switching without process restart

33 Tools. One Connection.

Every tool maps to a named Percepxion API endpoint. Your AI sees the tool description and calls it by name, you see the result.

Running in Under Five Minutes

Works with Claude Desktop, Claude Code, GitHub Copilot, or any MCP-compatible client on macOS, Linux, or Windows with WSL.

F A Q

Common Questions