Skip to main content

"So… what's the ROI?" How I automated Copilot business cases with MCP

Copilot Consultant MCP icon

Every Microsoft 365 Copilot conversation I have with a client ends with the same question: "So… what's the ROI?"

For a long time my answer was a spreadsheet, a few screenshots and a slide deck I rebuilt by hand for every client. It took half a day, and the numbers were only as consistent as my copy-and-paste.

AI assistants can help, and I use Copilot Chat, Researcher and Cowork skills. But when I asked them for an ROI dashboard, I got a slightly different formula and a different layout each time. That's fine for a one-off. It isn't fine for a number a CFO will question.

So I built my own Model Context Protocol (MCP) server: the Copilot Consultant MCP. It gives an AI assistant a set of consulting tools (readiness, build path, use cases and ROI), plus tools that produce a client-ready dashboard and PowerPoint deck. The maths lives in Python code, so the same inputs always give the same answer.

What the Copilot Consultant MCP does

  • Scores Copilot readiness out of 100 from five yes/no questions (licences, data governance, SharePoint permissions, executive sponsor, change plan) and lists the gaps.
  • Recommends a build path: Agent Builder, Copilot Studio or Azure AI Foundry, based on whether you need code, business-system connections or multi-step workflows. It also suggests use cases and plans a Build-Along workshop.
  • Calculates ROI with Microsoft's Copilot assisted hours method, in US$ or S$, with payback, break-even and a Forrester benchmark.
  • Reads real usage data from a Copilot Dashboard export, or downloads Microsoft 365 Copilot usage reports from Microsoft Graph.
  • Builds the deliverables: an interactive HTML dashboard and a PowerPoint deck with native, editable charts.

It has 10 tools, 2 reference resources and a discovery-call prompt. I use it in Claude Desktop, and the repo has setup steps for Claude Code, the OpenAI Codex app, ChatGPT (developer mode), Copilot Studio and Copilot Cowork.

Running inside Claude Desktop

On my Windows laptop, Claude Desktop starts the server itself, so there's no hosting and no URL. Once it shows as Running, I just talk to Claude and it picks the right tool.

Claude Desktop settings showing the copilot-consultant MCP server running
Claude Desktop starts the MCP server on my PC.
Claude listing the copilot-consultant tools
Asking "What tools does copilot-consultant have?"

Discovery and planning

This is where most of my client conversations start. In one prompt, I describe the client and Claude calls four tools (readiness, tier, use cases and Build-Along plan) and writes a meeting brief.

I'm meeting a Retail client's Operations team next week.
They have Copilot licences and an exec sponsor, no change plan,
DLP in place, SharePoint not reviewed. They want an agent that
checks inventory in their ERP and raises supplier requests.
Assess readiness, recommend a tier, suggest use cases and plan
the Build-Along.
Meeting brief showing the readiness score and recommended tier
The meeting brief: readiness score, gaps and the recommended tier.
Build-Along workshop plan expanded into a 90-minute session
A Build-Along plan for a Copilot Studio session.

Claude can also turn the readiness result into a visual. This infographic was made by Claude in the chat from the tool's output, not by the MCP itself:

Readiness infographic created by Claude from the assess_readiness result
A readiness infographic Claude created from the tool result.

ROI, the Microsoft way

The value side follows Microsoft's published Copilot assisted hours method from the Copilot Dashboard in Viva Insights: meeting hours summarised or recapped, plus 6 minutes per search or summary action, plus 6 minutes per creation action. Assisted hours × hourly cost gives the assisted value.

On top of that, the MCP adds the business-case steps: licence and rollout costs, year-one and 3-year ROI, payback, break-even minutes per user, and a comparison with Forrester's projected Total Economic Impact study for SMB (commissioned by Microsoft). It works in US$ or S$, and it asks for the hourly cost, licence price and rollout costs instead of guessing them.

ROI estimate in Singapore dollars with net benefit, ROI and break-even
An ROI estimate in S$ with the break-even point.

The interactive dashboard

One prompt creates a single HTML file with three tabs. The client can move sliders, switch scenarios and currency, and upload their logo. Nothing needs a server; it's just a file you can email or open in a browser.

Dashboard Business case tab with net benefit, ROI gauge and sliders
Business case tab: net benefit, ROI gauge against the benchmark, and live assumptions.
Dashboard Usage and adoption tab with weekly charts and adoption by organisation
Usage & adoption tab, filled from a Copilot Dashboard export (synthetic data here).
Dashboard Method and sources tab explaining the calculation
Method & sources tab, so every number can be traced.

Real usage data: export or Microsoft Graph

There are two ways to feed in real numbers:

  • Copilot Dashboard export. In Viva Insights, Export data > Export by week gives six months of per-person activity. The MCP matches Microsoft's metric names and calculates assisted hours per week, organisation and function.
  • Microsoft Graph Copilot usage reports (v2). The download_copilot_usage tool downloads the user detail, summary and trend reports to your Downloads folder. An admin signs in in the browser with their own account and MFA. The app registration uses the delegated Reports.Read.All permission, so no client secret is stored on the PC.
Claude summarising the synthetic Copilot Dashboard export
Summarising a Copilot Dashboard export (synthetic sample data).

Graph gives adoption, prompts and active days, but not meeting hours or creation actions. So from Graph alone, the tool reports a prompt-based lower bound and asks for the rest.

The client deck

The create_client_deck tool builds a PowerPoint deck: title, clickable agenda, readiness, roadmap, use cases, business case, value over time, usage, Build-Along and a method slide. Charts are native PowerPoint charts, so you can select one and choose Edit Data. Every slide has speaker notes, and you can give it your own widescreen template. There's an example in the repo: Sample_Copilot_Deck_Contoso.pptx.

Why an MCP, and not just Skills or Researcher?

You absolutely can get an HTML dashboard from Copilot Chat or Researcher today. The difference is trust and repeatability. Skills teach the AI how; MCP gives it hands.

  • Same inputs, same answer. The ROI maths is in code, not re-derived by the model on each run.
  • Same deliverable for every client. The same dashboard and editable deck, not a new layout each time.
  • Works beyond Microsoft 365. Claude, ChatGPT and Copilot Studio too, including in pre-sales before a client has Copilot.
  • Less data sent to the AI. Exports are processed on your PC, and only a summary goes to the model.

And it isn't either/or. The Copilot Cowork package in the repo is literally a skill plus this MCP: Researcher for the story, MCP for the numbers.

Getting started

What you need: Windows, Python and Git, plus Claude Desktop (or another MCP app). For the Graph download, you also need an admin in a tenant with Copilot licences.

  1. Clone the repo and install the libraries (PowerShell):
cd C:\Users\<you>
git clone https://github.com/shenukaf69/Custom_MCP.git
cd Custom_MCP
python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python demo.py
  1. python demo.py prints sample results and opens a sample dashboard, with no AI app needed.
  2. In Claude Desktop, open Settings > Developer > Edit Config and add the server to claude_desktop_config.json:
"mcpServers": {
  "copilot-consultant": {
    "command": "C:\\Users\\<you>\\Custom_MCP\\.venv\\Scripts\\python.exe",
    "args": ["C:\\Users\\<you>\\Custom_MCP\\server.py"]
  }
}
  1. Fully quit Claude (tray icon > Quit) and reopen it. The server should show as Running.
  2. Ask: "What tools does copilot-consultant have?" Then try the prompts in TRY_IT.md.

For ChatGPT, Copilot Studio or Copilot Cowork, the server runs as a web service (python server.py --http) behind a public HTTPS address, such as a Dev Tunnel or Azure Container Apps. The steps are in INSTALL.md and DEPLOYMENT.md. Agent Builder can't add MCP servers, so for Microsoft 365 Copilot the route is a Copilot Studio agent or a declarative agent.

A few honest notes

  • What I've tested: the discovery, ROI, export and dashboard tools in Claude Desktop on Windows (the screenshots above are real runs), the deck through the demo script, and the web-service mode in a local test.
  • Not yet tested end to end: the hosted setups in Copilot Studio, Cowork and ChatGPT. The steps are documented from Microsoft's and OpenAI's docs, but I haven't run them in a live tenant yet.
  • The Graph download has been tested against a simulated Graph service, not a real tenant. Its first real run will be the real test.
  • Assisted hours are an estimate. Microsoft calls them directional, not a guaranteed saving. The Forrester figures are a projected study commissioned by Microsoft.
  • Data requirements: the Copilot Dashboard export is a preview feature and needs at least 50 Copilot or Viva Insights licences. The Graph reports cover licensed users only.
  • Fixed values: the exchange rate (1 USD = 1.28 SGD), Microsoft's factors and the benchmark are in the code and don't update themselves.
  • Be careful with client data. Whatever a tool returns is sent to your AI provider. The per-user CSVs stay on your PC, but only use real client data where your agreements allow it. If you host the server, protect it with an API key or OAuth.
  • Sample data is synthetic. "Demo Client" and "Contoso" are made up.
Get the Copilot Consultant MCP
📦 GitHub repo: github.com/shenukaf69/Custom_MCP
⬇️ Download: ZIP of the latest version (Releases page; no tagged releases yet)
📖 Docs: README · Install guide · Try it · Graph setup · Hosting

If this is useful, a ⭐ on GitHub or a comment with feedback would mean a lot. What would you automate first?

The Copilot Consultant MCP is an independent project by Shenuka Fernando. It is not an official Microsoft product and is not endorsed by Microsoft. Microsoft 365, Microsoft Copilot, Copilot Studio, Viva Insights, Microsoft Graph, Microsoft Entra, PowerPoint and Azure are trademarks of the Microsoft group of companies. Claude is a trademark of Anthropic, and ChatGPT and Codex are trademarks of OpenAI. Other names may be trademarks of their respective owners.

Comments

Post a Comment

Popular posts from this blog

Microsoft power apps to generate day end email stats

  Power automate to generate a report to get total incoming outgoing emails in my mailbox for the day You can use Microsoft Power Automate (formerly Microsoft Flow) to generate a report of the total number of incoming and outgoing emails in your mailbox for a specific day. Here's a basic example of how you can get started: Create a new flow: Start by creating a new flow in Power Automate. You can choose a template or start from scratch, depending on your preference. Connect to your email account: Next, connect Power Automate to your email account. You can use the Office 365 Outlook Connector or the Exchange Connector, depending on which email service you use. Trigger the flow: Choose a trigger for the flow that will run it automatically on a schedule. For example, you could use the "Recurrence" trigger to run the flow every day at a specific time. Count the emails: Use the "Get email" action to retrieve all the emails in your inbox for the specific day. Then, us...
Post a Comment
Read more

Microsoft Viva Insights and Viva Insights Manager app

  Microsoft Viva Insights is a platform designed to help organizations promote employee well-being and productivity by providing data-driven insights and recommendations. It is an AI-powered tool that provides personalized insights to individuals, managers, and leaders. Viva Insights uses data from multiple sources such as Office 365, LinkedIn, and other business applications to analyze work patterns and provide insights on employee well-being, productivity, and collaboration. The platform is designed to help individuals manage their work-life balance by providing insights and recommendations on managing work hours, prioritizing tasks, and taking breaks. It also enables managers to identify potential burnout risks, track employee engagement, and support their team's well-being. Viva Insights provides a range of reports and dashboards that allow organizations to track employee well-being and productivity metrics and make data-driven decisions to support their employees. Additionally...
Post a Comment
Read more

Microsoft Copilot: Your AI-powered Productivity Partner

Microsoft Copilot is not just a name; it's a vision for the future of work. It's a suite of AI-powered tools designed to revolutionize the way we work, making us more efficient, creative, and productive. Copilot isn't just one tool; it's an ever-evolving ecosystem of features and functionalities that are constantly being updated and improved. What is Microsoft Copilot? At its core, Microsoft Copilot is a large language model (LLM) chatbot developed by Microsoft. It was initially launched as Bing Chat in February 2023 and has since evolved into a more comprehensive suite of tools integrated into various Microsoft products and services. Copilot leverages the power of LLMs to understand your needs and provide assistance in various ways, from simple tasks like answering questions to more complex tasks like generating creative content or summarizing documents. Copilot in Action Here are some of the ways Copilot can be used to enhance your productivity: Microsoft 365 Copilot:...
Post a Comment
Read more