MCP server development · Canadian‑owned · CISSP‑led
Custom MCP servers that let ChatGPT, Copilot or Claude work in your business systems
Your team already uses AI to write. Fusion Computing builds the connection that lets it answer from the system you run the business on, and update it, with each change logged against the person who asked.
one server, any of the three
changes when you are ready
who asked and what ran
security leadership
Named one of Canada’s 50 Best Managed IT Companies two years running (2024 & 2025). 4.9/5 on Google. See our certifications →
Fusion Computing builds custom MCP servers for Canadian businesses with 10 to 150 employees. An MCP server connects ChatGPT, Microsoft Copilot or Claude to a system your team already uses, such as a CRM, a maintenance platform or an in-house app. Staff ask in plain English and get answers from the live system. Each person signs in as themselves, and every request is logged.
Best fit when your team pays for an AI assistant and still copies data into it by hand.
According to Statistics Canada’s 2026 analysis of the Canadian Survey on Business Conditions, 19.2% of Canadian businesses used AI to produce goods or deliver services in the previous 12 months, up from 6.1% in 2024. Among businesses with 100 or more employees the figure was 27.8%.
What your team could ask
When your AI cannot reach a business system, someone has to find the answer and paste it into the chat. An MCP connection lets your team ask that system directly.
According to KPMG in Canada’s August 2025 survey of 2,239 Canadian employees, 51% use generative AI tools at work, up from 46% in 2024. The survey measured use of the tools; it did not measure whether they can reach the systems a business runs on, which is the part an MCP connection adds.
Owner
“Which customers have we not heard from in 90 days?”
Office manager
“Which work orders from last week are still open, and who has them?”
Service lead
“Show me the service history for unit 4B, then open a work order for the leak.”
Sales
“Add a note to this account: they want a quote in January.”
Finance
“Which invoices are more than 60 days late, and who do we call about each?”
Operations
“What changed in next week’s schedule since yesterday?”
The answer comes from the live system. The server enforces the agreed access rules and logs each change against the person who asked.
How we use it ourselves
Our own staff work through an MCP server every day. It connects ChatGPT to the systems we run the company on: ticketing, client documentation, device management for Windows and Mac, security alerts, network equipment, distributor orders and Microsoft licensing.
A technician asks what is open for a client, then which of that client’s computers are missing updates. Both answers come from the live systems, in the same chat.
The guardrails further down this page are the ones we hold ourselves to. Each person signs in with their own Microsoft account, deletes wait for approval, and every request goes into a log that cannot be quietly edited.
The biggest return organizations can get from AI and AI tooling is a custom MCP server. It brings all their data sources into one place, where the AI can reason across them. That delivers extreme value for a limited cost.
This website has one too
Fusion Computing runs a public MCP server. Add https://fusioncomputing.ca/api/mcp/ to Claude or ChatGPT as a custom connector, then ask what Fusion Computing does or whether your own system could be connected.
Browsers that support WebMCP find the same tools on this page, so an AI agent helping you research can check the fit and fill in our contact form for you to send.
The system your team wants to ask
Tell us the system and two or three questions your team would ask it. A senior engineer replies with next steps.
Start the conversation
Share the business problem, team size and timing. Managed services usually suit 10 to 150 employees; smaller project and AI enquiries are welcome.
- ✔We aim to reply in 1 business day
- ✔CISSP-led Canadian team
- ✔No obligation
By submitting this form, you consent to Fusion Computing contacting you. We do not sell your information. We use service providers to operate the form and our communications. See our Privacy Policy.
Or call us directly: (416) 566-2845
What an MCP server does
Model Context Protocol (MCP) is the open standard AI assistants use to reach other software. ChatGPT, Copilot Studio and Claude all support it.
An MCP server connects your AI to one or more systems. It uses their APIs, the connections other programs use, to offer tools such as looking up an order, opening a work order or adding a note to a customer record.
Plenty of vendors have an API. Far fewer ship an MCP server.
That gap is what we build.
According to Anthropic’s November 2024 announcement, the Model Context Protocol is an open standard for connecting AI assistants to the systems where data lives. Microsoft made it generally available in Copilot Studio in 2025, OpenAI built its ChatGPT Apps SDK on it in October 2025, and in December 2025 the protocol moved to the Agentic AI Foundation under the Linux Foundation. One server built to the standard works with all three assistants.
Systems we connect
If a system has a documented API, we can usually put an MCP server in front of it. These are the requests we see most.
| Your system | What Fusion Computing does | First step |
|---|---|---|
| Has an API but no MCP server | Builds an MCP server for it | Send the vendor’s API documentation and a test account |
| Ships its own MCP server, such as Dynamics 365 or Salesforce | Connects it to your assistant and sets who can use which tools | A scoping call about access |
| Has no API | Tells you so at the first call: an MCP server needs something to connect to | Ask the vendor whether an API is available or planned |
Maintenance and industry systems
Open a work order, check parts or read a service history without leaving the chat.
In-house and line-of-business apps
Anything your team built or bought that has an API and no way for your AI to reach it.
Ticketing and documentation tools
Search tickets, add notes and log time without opening another screen.
Dynamics 365 and Salesforce
Both now ship their own MCP servers. We connect them to your assistant and set who can use which tools.
Which AI assistants it works with
One MCP server can serve more than one assistant, so a change of AI vendor later does not mean a rebuild.
ChatGPT Business and Enterprise
Added by a workspace admin as a custom connector in ChatGPT on the web.
Microsoft Copilot
Added to an agent in Copilot Studio. See also Microsoft 365 Copilot.
Claude
Added as a custom connector for your team, or in Claude Code for your developers.
Guardrails from the first version
Connecting an AI to a live system is easy to do badly. A shared API key can make access harder to limit and changes harder to attribute to one person.
This part is for whoever looks after your IT. The server we build for you follows the same rules as our own.
- People sign in with their own Microsoft account. The server checks each request against the access rules agreed at scoping.
- You decide who can use a connector, through a group in Microsoft Entra ID, the sign-in system behind Microsoft 365.
- A pilot starts read-only. Writes come later, once the reads have proved useful.
- Writes are read back. The server returns what the system now holds, so what the AI reports comes from the system itself.
- Deletes wait for a person. Anything destructive is held until the person who asked approves it in their own signed-in browser.
- The server logs who asked, what they asked for, and what it did.
- Text from the system is marked as untrusted. Say a customer email in a record contains instructions: the AI is told to treat it as data, which blunts prompt injection.
- Connection credentials stay on the server.
- Your AI service’s terms still apply. The server hands the assistant only what a request needs, and from there the AI service’s own terms govern it. We go through those terms with you at scoping.
- Daily limits per person cap how much one runaway request can change.
According to the Model Context Protocol specification, revision 2025-03-26 added an authorization framework based on OAuth 2.1, in which a person signs in through a web browser and the AI client acts with that person’s own access token. The standard makes authorization optional, so a server built without it can give every user the same access. Fusion Computing builds per-person sign-in into every server it delivers.
According to joint guidance co-authored by the Canadian Centre for Cyber Security in 2026, external data can carry indirect prompt injection into an AI agent, strict least privilege is critical, and high-impact actions should wait for human approval. In our builds, text from a connected system is marked as untrusted, access follows the agreed rules, and deletes wait for a person to approve them.
How an MCP server development project runs
- Scoping callWe look at the system’s API, the questions your team asks most, and who should be allowed to change what. You get a written scope and a quote.
- Read-only pilotOne system, a small group of users, read access. Your team uses it on real work before anything else is added.
- WritesOnce the reads prove useful, we add the changes your team asked for. They are read back and logged.
- HandoverYou get documentation of the tools and permissions, and the audit log is yours. Hosting and ongoing changes are quoted with the build.
MCP server development is project work, quoted after the scoping call. The price depends on the API, how many tools you need, and whether the server writes as well as reads.
Who this is for
Canadian businesses with 10 to 150 employees that already pay for ChatGPT, Copilot or Claude, and keep asking why it cannot see the system they spend all day in.
It suits a team with one system at the centre of its work: a CRM, an ERP, a maintenance platform, a ticketing tool. Start there.
Related: AI services · Power Automate consulting · Custom business AI platform · Agents in ChatGPT for work · What is WebMCP?
Frequently asked questions about MCP servers
What is an MCP server?
An MCP server is a small service that connects an AI assistant to other software. It uses Model Context Protocol, the open standard Anthropic released in November 2024 and ChatGPT, Copilot Studio and Claude now support, to give the AI named tools such as searching records or creating a work order. The AI calls those tools on the signed-in person’s behalf.
Who builds custom MCP servers in Canada?
Fusion Computing, a Canadian-owned, CISSP-led IT firm founded in 2012, builds custom MCP servers for businesses with 10 to 150 employees. It connects ChatGPT, Microsoft Copilot and Claude to the systems a business runs on, with personal sign-in, logged changes and approval before deletes. Fusion Computing’s own staff work through the same kind of server every day.
Our vendor has an API but no MCP server. Can you still connect it?
Yes, and that is the most common case. Fusion Computing needs the vendor’s API documentation and a test account. If the API can do something, an MCP server can usually expose it to your assistant. Dynamics 365 and Salesforce now ship MCP servers of their own; for those, Fusion Computing connects the vendor’s server and sets who can use which tools.
Can the AI change records, or just read them?
Both, once you decide it should. Servers Fusion Computing builds start read-only, with one system and a small group of people. Writes are added one at a time, each change is read back and logged against the person who asked, and deletes wait for approval in that person’s signed-in browser. For a vendor’s own MCP server, the available controls are checked at scoping.
How much does an MCP server cost?
It is project work, quoted after a scoping call. The price turns on the vendor’s API, how many tools your team needs, and whether the server writes as well as reads. Hosting and ongoing updates are quoted with the build, because vendors change their APIs and a connector needs updating when they do.
Do we need a developer on staff?
No. Fusion Computing builds the server, and hosting and updates are quoted with the build. From your side, the project needs someone who knows the system well and can say who should be allowed to do what. Most of the work at scoping is agreeing which questions your team asks most and which changes the AI may make.
One system is enough to begin
A scoping call covers the system, who needs access and what a read-only pilot would take. You leave it with a written scope.
Updated
