Business Case MCP Server

Ask a chat assistant what an SEO recommendation is worth. It works through seven steps, from where a metric stands today to what moving it returns in money, and it never fills in a number you did not give it.

Add It To Your Assistant

  • Claude
  • ChatGPT
  • Gemini
  • Grok
  • Perplexity
  • Le Chat
  • Meta Muse
  • Manus
  • Copilot
# Connector address
https://mcp.jakelabate.com/business-case
# Then ask
Help me work out what an SEO recommendation is worth and whether it pays back. Ask me for the numbers one step at a time, and do not fill in any I have not given.

No sign-in. It works from the numbers you give it, reads nothing and keeps nothing.

7

Steps

From where a metric stands to what moving it returns.

8

Uplift methods

Ways to estimate an uplift, each a formula on show.

3

Business models

Kinds of business it can put a value on a click for.

5

Tools

None of them changes or stores anything.

How To Add It

  1. In Claude, open Customize, then Connectors, choose Add custom connector and paste the address. Claude's own instructions have the details.
  2. In any other assistant that supports MCP, add the address as a remote MCP server.
  3. Ask it about a recommendation. The prompt in the box works as it is copied.

What To Ask

  • "How do I justify this recommendation to the client?" It returns the seven steps, the question each one asks, and what makes an answer sound. get_framework
  • "These 40 queries should move from page two to the top five. What is that in clicks?" It works out the clicks now and after, from your own click-through curve, as a range. estimate_uplift
  • "What is one organic visit worth to this store?" It works it out from conversion rate, order value and margin, and says whether it is revenue or profit. value_per_unit
  • "Is this fix worth 60 hours of developer time?" It puts the seven steps together and returns the range of return, the arithmetic, and the certainty at which it breaks even. size_recommendation
  • "I do not know the conversion rate yet. What can you tell me?" It names what is missing and the question to ask for each. It does not fill the gap with a typical figure. size_recommendation
  • "Which of these six recommendations should we do first?" It sizes each one, ranks them on the cautious end of the return, and names the pairs the numbers cannot separate. rank_recommendations

What It Returns

Review stars on product pages: 20,000 clicks a month, expected to reach 22,000 to 24,000, at 0.72 USD of profit a click, for 40 to 60 hours of work. Sent to size_recommendation, this is the server's own answer, run when this page was built.

size_recommendation.jsonJSON
{
  "return": {
    "low": 360,
    "high": 16368
  },
  "currency": "USD",
  "period_months": 12,
  "pays_back": "at both ends of the range",
  "arithmetic": [
    "Timing: nothing in month 1, then building in a straight line to full effect over 2 months. Over 12 months that adds up to 10.5 months at full effect.",
    "B = 20,000 x 12 months = 240,000 clicks",
    "U = 240,000 + (22,000 to 24,000 - 20,000) x 10.5 = 261,000 to 282,000 clicks",
    "(U - B) x V = 21,000 to 42,000 clicks x 0.72 USD = 15,120 to 30,240 USD",
    "x C = x 50 to 70% = 7,560 to 21,168 USD",
    "W = 40 to 60 hours x 120 USD an hour = 4,800 to 7,200 USD",
    "R = 360 to 16,368 USD. The low end takes the low uplift, the low certainty and the high work. The high end takes the high uplift, the high certainty and the low work."
  ],
  "break_even": "To pay for itself within the period: on the cautious figures (low uplift, high work) it has to be at least 47.6% likely to work. On the hopeful figures (high uplift, low work) it has to be at least 15.9% likely to work. The certainty given is 50 to 70%."
}

The answer is a range because three of its inputs are. A single figure would hide how much of it is judgment.

The Seven Steps

Six questions to answer and one result. Each step is a number, or a claim, that someone can be asked to stand behind.

  1. Baseline (B). Where the metric stands now, with nothing changed. What is the number today, and where does it come from?
  2. Rationale. Which metric moves, and why this change moves it. What exactly is claimed to happen, and what makes you think so?
  3. Uplift (U). Where the metric would stand once the change has full effect. A range, low to high. What would the number be with the change in place, at the cautious end and at the hopeful end?
  4. Value (V). What one unit of the metric is worth in money. What is one more click, lead or order worth to the business?
  5. Certainty (C). How likely the uplift is to arrive, from 0 to 100%. A range. How sure are you, and what is that based on?
  6. Work (W). What it costs to do: hours and hard costs, as a range, and when it lands (months until it is live, months from live to full effect). What does it take to ship, and how long until it is live and working in full?
  7. Return (R). What is left over the period: the gain in money, weighted by certainty, less the work. Is it worth doing, and how does it compare with the other things that could be done?

The steps fall into three groups.

  • The claim: Rationale. What is being asserted. The other six steps are what it is worth if the claim holds.
  • The benefit chain: Baseline, Uplift, Value. From where the metric is, to where it would be, to what that difference is worth.
  • The judgment: Work, Certainty, Return. What it costs, how sure it is, and what is left.

The Formula

The formulaText
R = (U - B) x V x C - W

Return = (Uplift - Baseline)
         x Value x Certainty - Work
  • U - B is the gain. Uplift is the new level of the metric, not the difference. Taking the baseline away leaves what the change added.
  • x V turns it into money. Value is the worth of one unit, so it has to be counted in the same unit as the baseline and the uplift.
  • x C weighs it by how sure it is. A gain that is half likely counts for half.
  • - W takes away the cost. Hours at their rate, plus anything bought.

Baseline and Uplift are given for one month. The period, and the timing that comes with Work, turn them into totals: nothing counts until the change is live, and from then the gain builds in a straight line to full effect. The formula is worked on those totals, so it needs no term for time, and nothing is projected beyond the period you name, up to 120 months.

How Uplift Is Estimated

8 small models, each for one kind of change. Each takes measured numbers from your own data, shows its arithmetic, and returns the baseline and the uplift for one month in the same unit.

  • Ranking change. Queries are expected to move to a better position. For each query: searches x (CTR at the target position - CTR at the current position). The baseline is searches x CTR at the current position. ranking_change
  • Click-through change. Titles, descriptions or rich results are expected to earn more clicks at the same position. Baseline = impressions x current CTR. Uplift = impressions x new CTR. ctr_change
  • Indexation. Pages that are not indexed are expected to get into the index. Uplift = pages x share expected to be indexed x clicks per indexed page. The baseline is 0: a page that is not indexed gets no clicks from search. indexation
  • New pages. New pages are to be published for demand the site does not cover. Uplift = pages x share expected to succeed x clicks per successful page. The baseline is 0: the pages do not exist yet. new_pages
  • Conversion change. The change makes existing visits convert better: speed, layout, page experience. Baseline = sessions x current conversion rate. Uplift = sessions x new conversion rate. conversion_change
  • Recovery. Clicks were lost, to decay, cannibalization or an update, and part is expected back. Uplift = clicks now + (clicks before - clicks now) x share expected to recover. The baseline is clicks now. recovery
  • Risk avoided. The work prevents a loss: a migration, a redirect map, a template change. Uplift = clicks at risk x chance of the loss x share lost if it happens. The baseline is 0 clicks protected. risk_avoided
  • Stated directly. The baseline and the uplift are already known, from a test or another model. Baseline and uplift are taken as given. direct

There is no built-in click-through curve and no typical rate of any kind. A ranking change is read off the curve from your own Search Console, because a borrowed curve is a guess with a citation.

How Value Is Worked Out

What one click from search is worth, for 3 kinds of business. For an uplift counted in conversions, the same tool gives the worth of one order or one lead.

  • Online store. Conversion rate x average order value. ecommerce
  • Lead generation. Lead rate x close rate x value of a customer. lead_gen
  • Advertising and media. Pageviews per visit x revenue per 1,000 pageviews / 1,000. media

With a margin the value is profit. Without one it is revenue, and the answer says so every time, because a return worked out on revenue overstates what the business keeps.

Ranges And Missing Numbers

  • A number that was not given is never filled in. The return stays empty, and what is missing is listed with the question to ask for it.
  • Uplift, Certainty and Work are ranges. The return is a range too, worked out at every combination of their ends.
  • Baseline and Uplift are in the same unit, and Value is the worth of one of that unit. When the units differ, no return is given.
  • Every number can carry where it came from and how it is known. The ones that carry neither are listed.
  • The period is named by the caller. Nothing is annualized or projected beyond it.

Beside the return, the answer gives the certainty at which the work breaks even, on the cautious figures and on the hopeful ones, and the month in which the work is covered. Both can be checked against the certainty you gave.

Each number can say how it is known:

  • Observed. Counted or measured directly, such as clicks in Search Console.
  • Derived. Worked out from observed numbers by arithmetic, such as clicks divided by impressions.
  • Estimated. Produced by a model, such as search volume or a click-through curve.
  • Claimed. Someone said so, and it has not been checked.
  • Forecast. A statement about the future. Every uplift is one.
  • Unknown. Not known. Better said than hidden.

How Recommendations Are Ranked

  • Each recommendation is sized the same way. Up to 50 at a time, with the same formula, so the order comes from the inputs and not from who argues best.
  • The order is by the cautious end of the return. What each one brings in if the low figures are the true ones. It can be ranked on the hopeful end, or on return for each unit of work, when hours are what is scarce.
  • Overlapping ranges are named. When two recommendations next to each other in the order have ranges that overlap, the numbers do not settle which comes first, and the answer says so.
  • What cannot be sized is listed apart. With what each one is missing. It is never ranked on a guess.

The 5 Tools

  • get_framework Returns the framework for sizing an SEO recommendation in money.
  • estimate_uplift Works out steps 1 and 3, Baseline and Uplift, for one month.
  • value_per_unit Works out step 4, Value.
  • size_recommendation Puts the seven steps together and works out the return in money over a named period.
  • rank_recommendations Sizes up to 50 recommendations with the same formula and puts them in order of return.

From A Range To A Decision

A sized recommendation tells you what the work is worth and how sure that is. Getting the inputs measured, the work shipped and the result checked against the estimate is what an engagement is for.