Skip to content

BDQ practice

Your service desk, in the AI assistant you already work in

Most people would rather ask for help than go looking for it. Connect your service desk to an AI assistant and they can: an employee types a question in their own words, and the assistant finds the knowledge base article, raises the request or checks on a ticket, without anyone opening the portal.

The connection is MCP (Model Context Protocol), an open standard. The service desk platform publishes what an assistant may read and do, and the assistant connects to it.

So far we've shown this with Claude and Freshservice, and that's what the demo below covers. Because MCP is an open standard, the same approach is designed to work with other assistants, such as Microsoft 365 Copilot and ChatGPT. Freshworks says Freshservice works with any MCP-compatible tool. We'll say "works with" on this page for each assistant once we've tested it ourselves.

See it in Freshservice

Igor, one of our consultants, recorded Claude working with Freshservice in our own demo instance.

  • 0:56 Asking the knowledge base: how to get a guest laptop onto the Wi-Fi.
  • 1:22 Requesting software: Claude already knows the person's devices, what they have and what's in progress, and it shows the details for them to confirm before it raises anything.
  • 2:07 Following up: asking what's happening with a ticket, in the same chat.

Questions to try:

  • "How do I connect a visitor's laptop to the guest Wi-Fi?"
  • "I need Premiere Pro on my laptop. Can you request it?"
  • "What's happening with my laptop request?"

How we work

Why we run BDQ Claude-first

We're a small consultancy doing work that, done properly, needs a much bigger team: our website, our analytics, our finance inbox, and the service desks we run for customers. We couldn't do it all well by hand, so we built ways for AI to do the repetitive parts, with a person checking what matters. Why we run BDQ Claude-first tells the longer story.

Asking

Chris used to read Google Search Console and Plausible reports and work out what they meant. Now he asks Claude in plain English: "how is the site doing?", or "is the Cherwell page catching anything?" Claude pulls the live figures through a read-only connection, explains them, and suggests changes with the evidence attached. Chris approves every change before it goes live. He's written about it: I stopped reading Search Console reports. Now I ask.

What went wrong along the way

In its first week, the analytics connection mislabelled some of our pages, which is why we checked its numbers against a known export before trusting them. Separately, the claims checks we now run on every page caught a statistic on our own homepage that wasn't true.

Working this way lets a small team do more than it could by hand. It's the same arrangement as in the demo above: Claude does the looking up, and nothing is raised until the person confirms.

Governance basics

Before switching this on, you'll want to know what the assistant can see, what it can change, who controls it and what it costs. This is how it works with Claude and Freshservice today.

It acts as the person, with their permissions

When Claude works in Freshservice, every request runs as the person who connected it. Their role, field-level permissions and data scoping apply exactly as they do in the Freshservice screens, so Claude can't see or do anything that person couldn't.

Employees asking for help, as in the demo, connect with OAuth only and get their usual requester permissions.

On Claude's side, connectors inherit each person's permissions from the connected service, and on Team and Enterprise plans each person still signs in individually after the connector is switched on.

Admins decide what's switched on

In Freshservice:

  • MCP is on by default, and an admin can switch it off in Admin Settings.
  • Admins choose which tools are available, and can set up separate configurations with different tools for different teams. A tool that isn't switched on can't be found or used by the assistant.
  • Access for a connected AI tool can be revoked.

In Claude, on Team and Enterprise plans:

  • An Owner has to enable a connector for the organisation before members can use it.
  • Owners can limit what a connected service may do across the organisation, for example allowing reads but not changes.

What it costs

Freshservice meters MCP use separately from agent licences. It's available on the Growth, Pro and Enterprise plans, every call counts as one action whether it reads or changes anything, and each account has an annual allowance that doesn't roll over. Admins are notified as usage approaches the limit, and extra action packs can be added. Current figures are on Freshworks' pricing page.

If you expect heavy use, check your allowance before rolling it out widely. Claude needs its own plan too, and the organisation-wide connector controls come with Team and Enterprise.

If you run Freshservice and want to see this working, we'll take you through the demo live and talk through what connecting yours would involve: which tools to switch on, who gets access, and what it would use of your MCP allowance.

Accuracy note: platform features, plans and licensing change often. We last reviewed this page on .

Book a 30-min consult

No deck. No sales pitch. Just a useful chat.

Tell us roughly what you’re facing. We’ll come back within one business day with thoughts, references, and a sense of whether we’re a good fit.

  • ✓ Free, no obligation
  • ✓ Direct line to a senior consultant
  • ✓ Honest assessment — we’ll say no if it’s not a fit

We respond within one business day. No spam, ever.

Let's talk

See it on your own service desk

We use cookies to keep the site working and to power features like our live chat. Analytics are cookieless. See our privacy policy.