Build your own AI assistant — MCP servers & Skills

The AI assistant built into YTS Terminal Manager started life as a copilot that reads your live terminal and explains what it sees. It has grown into something bigger: a platform you can extend. Connect it to your own tools and data with MCP servers, and teach it your team's procedures with reusable Skills — right inside the terminal you already work in all day.

A quick recap: an AI that speaks CLI

The assistant lives beside your sessions. It watches the output scrolling past, so you can highlight a wall of logs and get a plain-English explanation, ask it to draft the next command, or hand it a task and let it work against the tab in front of you. It runs on the model you choose — Anthropic Claude, OpenAI, Google Gemini, Qwen, GLM and GitHub Models, or a fully local model through Ollama, LM Studio or vLLM. You bring your own key; the assistant stays optional.

That's a capable copilot. But a copilot only knows what's on the screen. The two features below let it reach further — into your systems, and into your way of doing things.

Bring your own tools with MCP servers

MCP — the Model Context Protocol — is an open standard for connecting AI assistants to external tools and data sources. An MCP server is a small, self-contained connector that exposes one system to the assistant in a structured way. Point Terminal Manager's assistant at a server and it can act on that system directly, instead of you copy-pasting between windows.

Because MCP is a standard rather than a closed integration, the assistant can talk to any compatible server — one you found, or one you wrote yourself. Typical connectors a network or systems team might wire up:

• A device inventory or CMDB — so "what's the mgmt IP for the core switch in rack 4?" is answered from the source of truth, not memory.
• A configuration or change database — pull the last known-good config, or diff it against what's live.
• A ticketing / ITSM system — read the incident you're working, or open one with the evidence already attached.
• Your internal docs and runbooks — the assistant cites your wiki instead of guessing.
• A monitoring or metrics API — correlate what the terminal shows with what the graphs say.

MCP servers can run locally on your own machine or reach a service on your network, so the connection stays inside your environment. You decide which servers the assistant is allowed to use.

Teach it your workflows with Skills

A Skill is a packaged, reusable set of instructions that teaches the assistant how you do a particular job. Where an MCP server gives the AI new reach, a Skill gives it know-how — the steps, the guardrails and the house style for a task — and loads it on demand, only when it's relevant.

Instead of re-explaining a procedure every session, you capture it once:

• A switch- or firewall-upgrade runbook — pre-checks, the exact command order, and the rollback path.
• A triage playbook — the sequence you always run when a link flaps or a tunnel drops.
• A compliance check — confirm a device meets your hardening baseline before it goes live.
• Vendor-specific quirks — the little differences between Cisco, Juniper, MikroTik and RAD that trip people up.

Skills make the assistant's output consistent and repeatable across a whole team: the junior on call gets the same careful procedure the senior engineer wrote, every time.

Better together: reach plus know-how

The two features are most powerful combined. Ask the assistant to "upgrade the access switch in rack 4," and it can load your upgrade Skill for the procedure while an inventory MCP server supplies the device details and a config server hands over the current running config — then it walks the change with you, on the live session, one guarded step at a time. That's the difference between a chatbot and an assistant that actually knows your network.

Private by design

None of this weakens the privacy story that makes Terminal Manager usable in serious environments:

• Run the whole thing against a 100% local model (Ollama, LM Studio, vLLM) — nothing leaves your machine.
• MCP servers can run locally or on-prem, so your inventory and configs stay inside your network.
• The assistant is entirely optional, and a risk gate asks before any risky command runs.
• Credentials stay encrypted at rest with Windows DPAPI, exactly as before.

Try it

The assistant, MCP support and Skills ship inside YTS Terminal Manager for Windows 10 and 11 — currently free, with no license key or registration. Point it at your model, wire up a server or two, write your first Skill, and let the terminal start doing more of the work.