Olympus by Old Bones MCP Server for Claude and ChatGPT
By Old Bones Therapy
Overview
Olympus by Old Bones is business intelligence for Shopify stores. It brings your Shopify sales together with your ad platforms (Meta, Google, Reddit and TikTok), Google Analytics, Search Console and Klaviyo email, and works out what you keep after product costs, shipping, fees and ad spend. Each store's Olympus runs in a hosting account the store owns, so its data stays with the store. This connector lets Claude read your store's numbers and answer in plain words: how last week went, what changed, which products sell, whether customers come back, and what each channel earns. Every metric is defined once, so the answers match what you see in Olympus. Olympus's specialists each cover one area: Hermes the whole business and briefings, Ares paid ads, Aphrodite email, Demeter search, Hephaestus conversion, Athena products and customers, Plutus margins. Ask about any of them and Claude brings in the one who covers it, without needing their names. Each is also a prompt you can pick from Claude's + menu. It only reads. Olympus never changes your store, and requests pass through oldbones.ai to your Olympus without being stored. Each person connects their own Claude and signs in to Olympus as themselves. Requires a Shopify store with Olympus by Old Bones set up, running in the store's own Railway account, on the Pantheon plan (a 14-day free trial, then from $249 a month by store revenue). On the Free plan, Claude connects and says what Pantheon adds, without reading the store's numbers.
How to connect
- Claude: Open the Olympus by Old Bones listing in the Claude connectors directory and choose Connect. Claude listing Olympus by Old Bones exposes 30 tools here.
- ChatGPT: Open the Olympus by Old Bones listing in the ChatGPT app directory and choose Connect. ChatGPT listing Olympus by Old Bones exposes 30 tools here.
Tools
business_snapshot
ChatGPTconvene_council
ChatGPTdata_health
ChatGPTget_ad_changes
ChatGPTactor is null (unknown). Days before collection began may have Meta only (platforms on each day says which logs were read). Noise is left out: audience refreshes, image uploads, delivery pings, Meta's review steps. byDay lists each day's changes, oldest first, each with a one-line label.get_ads
ChatGPTreported is Meta's own count, which flatters view and engaged-view settings), its ad set's setting, and two different fatigue reads. wearingOut: link CTR down while frequency is up, the window's last 7 days against the 7 before: interest now. lifetime: times each person it reached has seen it since it first ran, and newReachShare, the % of its latest week's people who had never seen it, as of the newest week Meta measured whatever the window; worn names why an ad carrying real spend reads as worn, the fatigue flag's own rule, and band its lifetime frequency against the store's levels. A worn ad can still hold its clicks, and an ad wearing out can be new: say which question each answers. lowSample: too few purchases to rank on. Each ad's before is the same number of days just before the window (null if it didn't run then), so whether its CPA held up is in the same call; byWeek adds its weeks through the window (weeks, oldest first): use it with search or a small limit for one ad's trend, rather than a call per week. Each ad's destination is where it's set to send people (its link, the kind of page, redirectsTo when the store redirects it) and enhancementsOn, the Meta enhancements that can send its clicks elsewhere; landings is where its clicks actually landed, from the store's own sessions matched by the utm_content its URL tags fill in: share on its link, shares by kind of page, top pages with their orders and net sales, and landsElsewhere when most of them started somewhere else. landingsUnknown says why an ad's visits can't be told apart (no utm_content, or one naming the ad set) and how to fix it: never estimate its landings. Ads are sorted by spend and capped at limit; matched says how many fit. Google and Reddit ads aren't stored yet: get_campaigns for their campaigns. How Ares reads these is in get_specialist, agent "ares".get_bought_together
ChatGPTproduct for one product's partners. A pair first-time buyers make is a starter bundle; one returning customers make is a reorder. complete: false means a week kept only its top combinations, so counts are minimums. A pair with a note had one of its products as a free gift in some orders (a free pick over an order value, say): keep the pair where its orders put it and show the note as a short, quiet line under it, not a warning; paidOrders are the orders where both were paid for. giftsKnown: false means some weeks predate gifts being marked. How Athena reads these is in get_specialist, agent "athena".get_break_even
ChatGPTcm1Percent is contribution margin before ads, as Shopify sees it. coverage is how much of those sales had a product cost ("Cost per item") recorded, and missing the products sold without one, most first, with whether each has a cost now: a product sold without a cost counts as free, so CM1 reads high by its share. Check coverage before quoting any margin. lines are the break-evens: mer is the sales per ad dollar below which ads lose money on the first order; ncacFirstOrder the most a new customer can cost and still pay back on their first order; ncacFirstYear the same over their first year, from the store's own repeat buying (observed). Lines are null below minCoverage. now is the last 28 days' MER and nCAC to set against them. Every line is a best case: notIncluded lists the costs Shopify can't see. Beside cm1Percent, two margins, each for its own question: cm1OfSales (CM1 ÷ total sales, tax and shipping in) is what the lines are built on, for anything judged against ad revenue; cm1OfProductSales (CM1 ÷ product sales after discounts and returns, without tax or shipping) is the margin m a discount or a price change comes off: a discount d needs m ÷ (m − d) times the orders. Never do discount or price math on cm1Percent or cm1OfSales: they count shipping and tax as sales, so a discount looks dearer than it is. costBasis says what the store's Cost per item holds (landed, the unit price only, or not said): state it, never guess. Each figure is defined in the metric registry: get_metric_definition cm1_percent, mer or ncac. How Plutus reads these is in get_specialist, agent "plutus".get_campaigns
ChatGPTbefore). Each campaign's breakEvenRoas is from what its visitors bought (GA4, by UTM campaign), else the store's, and breakEvenFrom says which; leftAfterAds is what its revenue leaves at that bar after spend. changed: budget, bid and status changes in the window; budgetAt: whether its budget sits on the campaign or its ad sets, the level to judge it at; tooNew: spent under 1.5 times the target CPA or began under 3 days ago. A platform's settling days are still being credited: never call a trend on them. level adSet gives Meta's ad sets (the only platform that stores them), each held to its campaign's bar. Meta's ads one by one: get_ads. Use query_source only for what neither covers (keywords, placements). storeSales (the store's own sales) sits beside platformRevenue; rows can carry awareness, slowToCredit and, for ad sets, lowSample, and notes says what each flag in the result means. How Ares reads these is in get_specialist, agent "ares".get_current_data
ChatGPTget_customer_cohorts
ChatGPTsoFar has it to date. months is each month since the first: payingRate (who placed a paid order that month), and repeatRate and ltv to date. summary weights every finished cohort by its customers: the store's typical figure. An exchange, replacement or gift (a $0 order) is never counted as buying again, so repeat rates are lower than Shopify's own cohort report. Rows flagged lowSample have too few customers to read a rate from. The last 24 cohorts are kept. firstProducts answers which first product brings customers back: for each product (and Product type) a first order had, free gifts left out, the share of those customers who bought again within 3 and 12 months and that against the typical customer (vsTypical, 1.3 = 30% more came back); lowSample under 100 first orders. shopifyComparison is Shopify Analytics' returning customer rate over the last 90 days (of everyone who bought, the share who had bought before at any time), every order and paid only: a different question, so when someone asks why Shopify shows a higher rate, explain the difference with these figures. Each figure is defined in the metric registry: get_metric_definition repeat_rate_3m, repeat_rate_12m, ltv_3m or ltv_12m.get_daily_data
ChatGPTget_date_range
ChatGPTget_email
ChatGPTget_email_results
ChatGPTget_metric
ChatGPTget_metric_definition
ChatGPTget_organic_searches
ChatGPTget_page_funnel
ChatGPTchannels gives each channel's visits and conversion. Each rate is defined in the metric registry: get_metric_definition page_cart_rate, page_checkout_rate or page_conversion_rate.get_price_test
ChatGPTcontrol: every product whose price didn't move over those days, so traffic and the season net out (vsControl). Products that changed within 7 days of each other are one test: together is the change as a whole, and leads, since shoppers move between products that moved together. verdict (pays, costs, no clear change, too early, too few orders) is on CM1 a day against control and the product's own band, how far it moves in an ordinary month; rule states it. Quote why: the same figures every time. breakEvenUnits is the change in units at which the new price keeps the same margin. ga4 is buyers per view (all, new, returning) beside it: a rise is meant to cost some buyers, so it decides nothing; watch says when new visitors fell beyond their band. status final: the read won't change (28 days passed, or endedBy the next price change); provisional: more days are coming. overlapping lists the other timeline events in those days.get_product_funnel
ChatGPTgifts. A product with giftProduct: true was given free in the window: its cartRate includes gift picks and is not comparable; never rank it on cartRate. A product with trackingDrop set is one whose sales GA4 stopped seeing over the last 28 days while the store's held (broken tracking, or an add-on or bundle app selling it): its purchaseRate reads low, so never report it as a fall in sales; use Shopify's units for it. the product's change in purchase rate measured against every other product's, so a store-wide dip isn't blamed on one product. Products are grouped by their Shopify Product type (groups). GA4 sees most but not all buyers (blocked trackers), so compare a product with its own history, not with Shopify's order counts. Use by for a breakdown per channel, device, source/medium, visitor or campaign. Each rate is defined in the metric registry: get_metric_definition product_cart_rate, product_purchase_rate or product_revenue_per_view.get_product_margin
ChatGPTshipping: labels less what customers paid), Shopify Payments fees (fees) and ad spend (adSpend), each shared by its share of net sales (method: Olympus doesn't yet know which product an ad sold, so CM2 by product differs only by each product's own CM1). cm1 and cm1Percent: contribution margin before ads; cm2 and cm2Percent: after its share of ads. Units given away (a discount took their whole price) aren't their product's cost: giftUnits/giftCogs were free in an order that paid for something (a gift with purchase: shared by net sales as gifts, in CM1) and freeUnits/freeCogs in an order with nothing paid (free product for athletes, creators, promotions: marketing, shared as freeOrders with ad spend, in CM2). soldUnits/soldCogs are what it sold, and costShare is on those. Never call a product unprofitable for what it gave away. A product with noCost sold without any cost recorded: its margin is null, never estimate one. partialCost: some of its sales had none, so its margin reads high. coverage is the share of the window's sales with a cost; under minCoverage every margin reads high: say so first. costChecks lists products sold without a cost, costs that look wrong ($0 typed in, or at or above the price) and the highest cost shares. breakEvenRoas, for each product and each Product type in types: the platform ROAS below which ads on it lose money. Platforms count tax and shipping in revenue and net sales has neither, so it is net sales × totals.salesPerNetSale ÷ CM1, never 1 ÷ cm1Percent (that sets the bar too low). Judge an ad set that sells one type against its type's breakEvenRoas. The products' CM1 adds up to a little less than the store's (get_metric cm1): what free orders cost is marketing here, in CM2. Contribution margin is not profit: fixed costs come after it. How Plutus reads these is in get_specialist, agent "plutus".get_search_queries
ChatGPTmoved (negative is up), its topPage, branded or not, and flags: nearPageOne (positions 11 to 20, where a fix moves a search onto page one) and pageOneNoClicks. Filter with brand, search, page or near_page_one; sort by impressions, clicks, position or moved. From each day's top 300 search × page pairs: days says how many days a search was among them, and the notes say what share of impressions the rows carry. How Demeter reads these is in get_specialist, agent "demeter".get_search_terms
ChatGPTget_specialist
ChatGPTget_timeline
ChatGPTsource: "detected"): price changes (from what each variant sold at), launches, and products selling out or back in stock, each naming its product; a price change's detail says when sales only bracket the day it changed. Also page edits Olympus found (detected: "page") by comparing each product's and collection's page night to night: one event per page per day, its what (title, seo, copy, handle, status, products, rules, created, deleted), changes with before and after for titles, SEO titles and web addresses, and a weight. Major ones (a title, SEO title, web address or status) come by default; minor ones (a description, an SEO description, a collection's products or rules) only when asked with weight or what, and are counted otherwise. Before blaming Google or the ads for a page's drop, ask for its page edits, e.g. what "seo" with search "/collections/<handle>". The ad accounts' own change logs are get_ad_changes.get_weekly_data
ChatGPTlist_agents
ChatGPTlist_metrics
ChatGPTquery_source
ChatGPTreport_gap
ChatGPTwhoami
ChatGPTbusiness_snapshot
Claudeconvene_council
Claudedata_health
Claudeget_ad_changes
Claudeget_ads
Claudeget_bought_together
Claudeget_break_even
Claudeget_campaigns
Claudeget_current_data
Claudeget_customer_cohorts
Claudeget_daily_data
Claudeget_date_range
Claudeget_email
Claudeget_email_results
Claudeget_metric
Claudeget_metric_definition
Claudeget_organic_searches
Claudeget_page_funnel
Claudeget_price_test
Claudeget_product_funnel
Claudeget_product_margin
Claudeget_product_profit
Claudeget_search_queries
Claudeget_search_terms
Claudeget_specialist
Claudeget_timeline
Claudeget_weekly_data
Claudelist_agents
Claudelist_metrics
Claudequery_source
Claudewhoami
ClaudeLinks
Similar Apps
Listed on MCP App Store
Add this badge to your site or README. It links to this page.
HTML
<a href="https://mcp-app-store.keenethics-labs.com/app/olympus-by-old-bones"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>
Markdown
[](https://mcp-app-store.keenethics-labs.com/app/olympus-by-old-bones)





