This article is part of The MCP Guide, our plain-English reference on the Model Context Protocol.
An MCP business case is a short document that shows whether connecting your AI tools to your business systems through the Model Context Protocol is worth the money. It lists the work MCP would save, every cost it adds, and the point where the two break even. You write it before you fund a pilot or a build, so the decision rests on your numbers, not a vendor's.
This guide gives you a structure to copy, the cost lines people forget, and a worked example. Every number in the example is HYPOTHETICAL. It shows the method, not a result anyone achieved. As of November 2026, we know of no independent benchmark for MCP time savings that holds across companies, so we don't quote one.
If you have already started and want to measure real results, use our MCP ROI measurement playbook instead. This page is about the decision before that.
Key Takeaways
- A business case covers one workflow at a time, with a written baseline of hours, errors and current costs.
- Savings come from time, fewer errors and faster turnaround, but only time and errors are easy to price.
- Costs include AI seats, server or platform fees, setup, security review, training, review time and upkeep, not just a subscription.
- Test a low, middle and high scenario, then work out the break-even: the time reduction you need just to pay back the costs.
- Use the pilot to check the assumption with the most effect on the result before you spend on a full rollout.
What goes into an MCP business case
Keep it to two pages. A decision-maker should be able to read it in ten minutes. Use these sections:
- The workflow. One task, described the way people do it today. Example: "Account managers copy numbers from three tools into a weekly client report."
- The baseline. Hours per week, people involved, error or rework rate, and any tool or contractor cost today.
- The proposed change. Which AI app, which MCP servers, and what the AI may read or change.
- Savings. Time, errors and speed, each with the assumption behind it.
- Costs. Year one and year two, split into one-off and recurring.
- Scenarios and break-even. Low, middle and high, plus the break-even point.
- Risks and controls. What could go wrong and how you'll limit it.
- The ask. What you want approved now. Usually a pilot, not a rollout.
Where the savings come from
Time saved
This is the main line. Multiply the hours saved by a loaded hourly rate. A loaded rate includes wages plus benefits and overhead. Ask your finance lead for the figure your company uses, rather than guessing.
Be careful with vendor dashboards. Anthropic's usage analytics help article lists an "Estimated time saved" metric for Team and Enterprise plans, but it doesn't explain how that estimate is made. Use your own timings in the business case.
Fewer errors
Price errors by the time it takes to find and fix them, plus any direct cost such as a credit to a customer. If the AI can introduce new errors, count those too. A wrong record written by a tool still costs someone time to reverse.
Faster turnaround
Same-day answers or faster reports can matter a lot, but they are hard to price honestly. List them as unpriced benefits. Don't add a dollar figure you can't defend.
The cost lines people forget
The subscription is usually the smallest part. Here are the lines to include, with examples of published prices checked on 6 November 2026. Prices change, so open each page before you budget.
| Cost line | One-off or recurring | Where to get the number |
|---|---|---|
| AI app seats | Recurring | Your vendor's pricing page. Claude's pricing page lists Team standard seats at $20 per seat a month billed annually, or $25 billed monthly. |
| MCP server or platform fees | Recurring | The vendor. Zapier says Zapier MCP is included in every plan and "each tool call uses two tasks" from your quota. Composio's pricing page lists a $0 Hobby plan with 100,000 tool calls a month and a $29 Pro plan. |
| Hosting (custom servers only) | Recurring | Your host. Cloudflare's Workers pricing lists a $5 monthly minimum per account on the Paid plan. |
| Build (custom servers only) | One-off | A written quote. See what drives MCP server development cost. |
| Setup and configuration | One-off | Internal hours times the loaded rate. |
| Security review | One-off, then yearly | Internal or outside reviewer. Use our MCP security checklist to scope it. |
| Training | One-off | Hours per person times the loaded rate. |
| Human review of AI output | Recurring | Hours per week someone spends checking results. |
| Maintenance | Recurring | Hours per month for updates, broken connections and changed tools. |
Two lines deserve a note. Security review is real work, because the official MCP security best practices ask for a "progressive, least-privilege scope model" and forbid servers from accepting tokens not issued for them. Someone has to check that a server you rely on does this. Review time is the other one. The MCP tools specification says there "SHOULD always be a human in the loop with the ability to deny tool invocations." Approving and checking actions takes time, so cost it.
If you are deciding between a managed platform and a custom server, our build vs buy guide covers that choice. Token and usage costs per task are covered in AI agent cost per task.
A worked example (HYPOTHETICAL)
Everything in this section is invented to show the method. The company, hours, rates, prices and percentages are HYPOTHETICAL. They are not benchmarks, customer results or eSEOspace estimates. Replace every number with your own.
The HYPOTHETICAL setup
A small marketing agency wants AI to pull numbers from its analytics, ads and project tools into a weekly client report, using official MCP servers from those vendors. No custom build.
| Input (HYPOTHETICAL) | Value |
|---|---|
| Staff doing the task | 4 people |
| Time on the task today | 2.5 hours per person per week (10 hours a week) |
| Working weeks a year | 48 |
| Baseline hours a year | 480 |
| Loaded hourly rate | $55 |
| Report corrections today | 2 a month, 1.5 hours each (36 hours a year) |
| Assumed share of corrections avoided | Half (18 hours, or $990 a year) |
HYPOTHETICAL year-one costs
| Cost line (HYPOTHETICAL) | Year one |
|---|---|
| AI seats, 4 people | $1,200 |
| Server or platform fees | $600 |
| Setup, 20 internal hours | $1,100 |
| Outside security review | $1,500 |
| Training, 3 hours each for 4 people | $660 |
| Review of AI output, 0.5 hours a week | $1,320 |
| Maintenance, 2 hours a month | $1,320 |
| Total | $7,700 |
HYPOTHETICAL scenarios
The only big unknown is how much time the workflow really saves. So the example tests three assumed reductions. These percentages are assumptions to test, not expected outcomes.
| Scenario (HYPOTHETICAL) | Hours saved a year | Time value | Plus error savings | Minus costs | Year-one net |
|---|---|---|---|---|---|
| Low: 20% faster | 96 | $5,280 | $6,270 | $7,700 | -$1,430 |
| Middle: 40% faster | 192 | $10,560 | $11,550 | $7,700 | $3,850 |
| High: 60% faster | 288 | $15,840 | $16,830 | $7,700 | $9,130 |
HYPOTHETICAL break-even
Break-even is the time saving that just covers the costs. Here, costs minus error savings are $6,710. Divide by the $55 rate and you get 122 hours, which is about 25% of the 480 baseline hours.
So the decision rule in this example is simple. If a pilot can't show roughly a quarter of the time saved, year one doesn't pay back. Year two looks different. Setup, the security review and training drop out, so recurring costs fall to $4,440 and the middle scenario nets $7,110. A one-off yearly review would add some of that back.
What the example teaches
- The result swings from a loss to a gain on one assumption. That is the assumption your pilot must test.
- Review time and maintenance together cost more than the seats. Leaving them out would flatter the case.
- A custom build changes the picture. Add the quoted build cost as a one-off line and recalculate break-even.
Risks to put in the business case
Name the risks and the control for each. A decision-maker will ask.
- Too much access. Start read-only and add write tools later. See the official guidance above on least-privilege scopes.
- Wrong results. Keep a person approving writes and checking output during the pilot.
- Vendor change. Servers, tools and prices change. Note who watches for that.
- Data exposure. List which data the AI will see and confirm it is allowed under your policies.
- Adoption. If staff don't use it, no saving happens. Budget for training.
How to present it
Lead with the ask, then the break-even. "We want approval for a 30-day pilot. It pays back in year one if it saves at least 25% of the time on this task" is easier to approve than a long ROI table. Put the scenario table on page two.
Ask for a pilot first, with clear success measures. Our 30-day MCP pilot plan shows how to set the go/no-go rules so the pilot answers the break-even question.
Next step
If you want a second opinion on which workflow to pick and which costs apply to your systems, start with our MCP readiness audit. If the case points to a custom server, see eSEOspace MCP development; we scope and quote each project individually.
More in the MCP Guide
Latest research: State of MCP: October 2026
Frequently asked questions
Is there a standard ROI figure for MCP?
Not one we can source. As of November 2026, we know of no independent benchmark for MCP time savings that holds across companies. Build the case on your own baseline and test the key assumption in a pilot.
What loaded hourly rate should I use?
Use the figure your finance team already uses for staff costs, including benefits and overhead. If there isn't one, agree on a rate with finance before you start. Don't pick a rate that makes the case look better.
Should I include faster turnaround as a dollar saving?
Only if you can show the link to revenue or cost, such as a contract penalty you avoid. Otherwise, list it as an unpriced benefit. Decision-makers trust a case more when the soft benefits are kept separate.
How is a business case different from measuring ROI?
The business case comes first and uses estimates to decide whether to fund a pilot. Measuring ROI comes after, using real before-and-after data. Our measurement playbook covers that second step.
What if the low scenario loses money?
That is normal and useful to know. It tells you the minimum saving the pilot must show. If the pilot falls short of break-even, change the workflow or stop before you spend more.
