BEHIND THE CHAT · IT HELP DESK · TOOLS, MCP & CONNECTORS
What a help desk ticket costs
THE WORK FLOOR
🧭MODEL
Think / Plan
📚KNOWLEDGE
IT knowledge base
💻MCP SERVER
Device lookup (Intune)
✏️MODEL
Write reply
📣WORKFLOW
Alert IT in Teams
🎫CONNECTOR
Freshservice ticket
WHAT'S HAPPENING
THE BACKPACK
0
tokens the LLM re-reads on this model call (GitHub harness)
This lap's route Earlier lap Standard harness GitHub harness
COST REGISTERcredits · $0.01 each
STANDARD HARNESS0.00$0.00
GITHUB HARNESS0.00$0.00
Patch · IT help desk
Online
Type a message
➤
…
ONE PROBLEM · ONE TICKET
STANDARD HARNESS29.00credits
about $0.29
GITHUB HARNESS185.00credits
about $1.85
8 model calls · 1 knowledge search · 3 tools (MCP, connector, workflow)
Microsoft's guide for a light GitHub harness task: 100–300 credits. This one lands at 185.

How the register rings

Standard harness: a fixed rate card per eventTenant graph grounding (knowledge search) 10 · Generative answer 2 · Agent action (any tool: MCP, connector, workflow) 5 · Agent flow actions 13 per 100. Thinking is included. Included for Microsoft 365 Copilot licensed users in Microsoft 365 channels.
GitHub harness: usage-basedMicrosoft charges for LLM tokens, tools (including knowledge and MCPs) and the harness itself. Every model call re-reads the whole backpack, and tool results make it heavier: 30K → 49K tokens here. Never included in a Microsoft 365 Copilot license, and it bills while you build and test, too.
Ranges per GitHub harness task (from Microsoft's estimate chart)Light 100–300 · Medium 300–500 · Heavy 500+ credits.

Token math assumes $5 per 1M input tokens and $25 per 1M output tokens, 1 credit = $0.01. Microsoft doesn't publish separate harness and tool rates, so they aren't added here; a real meter can run higher. Ticket INC-4821 and all example savings are fictional.

CLASS IS IN SESSION
▶
Behind the chat about 8½ minutes
Press Play to start. Turn your sound on.
Patchyour help desk host
LibbyStandard harness
TurboGitHub harness

Behind the chat: what a help desk ticket costs

First, a quick class on AI models, tokens, credits and the two harnesses, with Libby and Turbo. Then watch Patch fix a problem, call an MCP server, open a ticket through a connector and alert IT with a workflow, while the register rings up both harnesses.