Quick answer: For a store an AI agent builds, the deciding question is not which tool is free but whether the free plan lets the agent read the numbers it needs. On their own pages, checked 2026-11-16, PostHog and Mixpanel are the two of these seven that describe all three of a permanent hosted free tier, funnels and a hosted MCP server an agent can connect to; Mixpanel adds that free plans can be subject to Query API access limits. If counts per event name are enough, Tolstoy's report gives them through its CLI or MCP at no charge, according to its docs, with no funnel or revenue. Orders and revenue still come from the store's own order data.
This comparison is for an AI agent that builds a small online store, and for the person who directs it. The agent picks an analytics tool before the store has traffic, usually on a "free" label. That label covers four different arrangements, and misreading one costs data or money later.
What "free" means, in four forms
- A hosted free tier with a cap. What happens at the cap matters: PostHog's pricing page says events past the free plan's limit are permanently dropped.
- A trial. Plausible's is 30 days with no card. Umami's 14-day trials are for its paid plans, separate from the free Hobby plan.
- Free software, paid hosting. Plausible Community Edition, Umami, GoatCounter and PostHog's open-source product cost nothing to license; the store pays for the server and its upkeep. Plausible's Community Edition leaves out funnels, user journeys, ecommerce revenue goals, SSO and the sites API.
- Free, without programmatic access. Umami's Cloud API keys, which its MCP server also uses, start on the Pro plan. Its free Hobby plan still documents the dashboard, CSV export by an emailed download link, and public share URLs. Whether an agent can use those paths depends on whether it can work a browser or read the email; the restriction is on the API, not on every way of reading.
The caps are in different units. Umami counts a pageview as one event and each stored event property as another; Plausible counts custom events toward billable pageviews; PostHog and Mixpanel count events.
The comparison, checked 2026-11-16
Each cell is the vendor's own statement on its pricing or documentation pages, listed in Sources. "Not verified" means the pages we read did not say; it is not a claim that the tool lacks it.
| Tool | What free covers | API or MCP read path |
|---|---|---|
| Tolstoy | Docs: events and the report free on every Tolstoy plan; no widget needed | Tolstoy CLI or MCP; sign-up and sign-in approval in a browser |
| PostHog | Hosted: 1M analytics events a month, resets monthly, no card; 1 project; events past the cap dropped. MIT open-source product | Hosted MCP, free to connect; some AI-powered tools billed as PostHog AI |
| Plausible | No free plan: 30-day trial. Community Edition free to self-host, without funnels or revenue goals | Stats API on Business. MCP: not verified |
| Google Analytics | Free of charge; 360 is the paid tier. User- and event-level data kept 2 or 14 months (explorations and funnels; standard reports unaffected) | Read-only MCP you run with a Google Cloud project; Data API quotas per standard property of 200,000 tokens a day each for Core, Realtime and Funnel requests |
| Umami | Cloud Hobby: 100K events a month, 1 website, 6-month retention. Self-hosting free (MIT) | Cloud API and MCP from Pro; self-hosted API, and MCP from v3.4.0 |
| GoatCounter | Hosted free for "reasonable public usage", donation-supported. Self-hostable | JSON API, 4 requests a second. MCP: not verified |
| Mixpanel | Free: up to 1M events a month, no card | Hosted MCP, on by default for free accounts created after August 1, 2026; free plans can face Query API access limits |
| Tool | Product view, add to cart | Funnels | Revenue | Freshness |
|---|---|---|---|---|
| Tolstoy | Names you choose | No | No; parameters not summed | Report from about 06:30 UTC next day |
| PostHog | Spec names (Product Viewed, Product Added) | Yes | Event property or Stripe source | Queryable within minutes, typically |
| Plausible | Custom events | Business plan | Business plan | Dashboard refresh every 30 seconds |
| Google Analytics | Recommended view_item, add_to_cart | Yes, up to 10 steps | Purchase events | Realtime in minutes; daily about 12 hours |
| Umami | Custom events | Hobby | Hobby; payment imports from Pro | Realtime view on every plan |
| GoatCounter | Named events, with title and referrer | Not described | Not described | Statistics update every 10 seconds |
| Mixpanel | Custom events | Free plan | Sum of a numeric property | Seconds after reaching ingestion servers |
How to read the tables
This is a descriptive reading of what the vendors' pages say on one date. We did not run each tool against a test store, and none of it ranks the tools.
The access column decides more than the price column. Six of the seven describe use at no charge, but they reach the agent in different ways: a hosted MCP (PostHog, Mixpanel), a CLI or MCP (Tolstoy), an MCP server the store runs with Google Cloud credentials (Google Analytics), a JSON API (GoatCounter), or, on Umami's free Hobby plan, no API at all, only the dashboard, exports and share links.
"Revenue" is as complete as the data supplied. The vendors document more than a browser script: Google's Measurement Protocol sends server-side and offline events to augment its tag, Mixpanel recommends tracking from the server where possible, PostHog can read revenue from a Stripe source, and Umami on Pro and Business imports payments and refunds from providers such as Stripe and WooCommerce. A revenue figure counts the purchases that reached the tool by any of these routes, which is why the store's order data stays the reference for revenue, AOV and refunds.
The freshness statements describe different layers. GoatCounter's statistics update every 10 seconds; Plausible's dashboard refreshes every 30 seconds; PostHog says events are typically usable in queries within a few minutes; Mixpanel's engineering blog (April 2024) says events are available within seconds of reaching its ingestion servers; Google describes processing intervals, with figures that can change for 24–48 hours; Tolstoy publishes a daily report. None of these is a guaranteed end-to-end delay from a shopper's click, and they are not equivalent promises.
How far to trust it. Plans and limits change, and each cell is a vendor statement, not a measurement. Check the current page before recommending a tool to a client.
Which tool fits which requirement
Counts of named storefront events, read by the agent. If the decision needs "how many shoppers viewed a product, and how many added to cart", per-name shopper counts are enough, and the agent computes step rates from them. Tolstoy and PostHog describe agent access at no charge; a self-hosted Umami offers its API and MCP if the store will run a server.
Funnels and session conversion. If the decision needs the same visitor to complete steps in order, use a tool that builds funnels. A ratio of two counts does not check that the same shoppers did both steps.
Privacy-focused traffic measurement. Plausible, Umami and GoatCounter each describe measurement without cookies or personal data, and Plausible and Umami say no consent banner is needed. Those are the vendors' statements, not a legal review. Consent handling also changes the denominator: the E-commerce Analytics SDK sends no event until the shopper allows analytics, except that with no consent tool on the page and no stored choice, it sends events.
Commerce revenue. Take orders, revenue, AOV and refunds from the store platform. An analytics tool's revenue view is for attributing sales to sources or steps.
Tolstoy's agent path, and its limits
The E-commerce Analytics SDK (the TolstoySDK package) runs in the storefront browser and sends events with names the store chooses. The report gives two numbers per event name for whole UTC days, up to 180 days: events, how many times it was sent, and shoppers, how many different shoppers sent it, with one shopper ID per browser. The docs state that sending events and reading the report are free on every Tolstoy plan, the free plan included, that no Tolstoy widget is needed, and that these events never count toward Shoppable Videos view billing.
There is no funnel, no conversion rate and no revenue for your events; parameters are kept but not shown, grouped or summed, so sending an order value does not create a revenue report. There are no numbers for today, no daily series and no server SDK, and a report lists up to 120 event names. Sign-up and the approval of the CLI or MCP sign-in happen in a browser, so an agent without one needs its person for those steps. Setup, consent and the same-day delivery check are in the Tolstoy e-commerce analytics guide, with reference in the Tolstoy SDK, Tolstoy CLI and Connect your agent pages; the ecommerce-analytics skill gives the agent an event plan and the formulas for rates from these counts.
Decide what to measure before choosing
A counts-only report cannot give conversion per session, and no analytics script replaces order data for AOV or refund rate. E-commerce analytics: what to measure and benchmarks by vertical sets out which metrics to track, their denominators, and median AOV and refund rate by vertical from Tolstoy's store data.
Sources
All checked 2026-11-16.
- Tolstoy: E-commerce Analytics SDK guide, ecommerce-analytics skill
- PostHog: pricing, capturing events, funnels, ecommerce events spec, revenue analytics, MCP
- Plausible: pricing, self-hosted Community Edition, custom events, funnels, revenue tracking, Stats API
- Google Analytics: Google Analytics, Analytics 360, data retention, measure ecommerce, Measurement Protocol, funnel exploration, data freshness, Analytics MCP server, Data API quotas
- Umami: pricing, track events, funnel, revenue, payment providers, Cloud MCP, self-hosted MCP, export data, share URL
- GoatCounter: home and pricing, events, FAQ, API
- Mixpanel: pricing, Insights, tracking methods, MCP server, infrastructure
Frequently asked questions
Which analytics tools are genuinely free for a small store?
Of the seven compared here, six describe use at no charge: PostHog, Mixpanel and Umami Cloud's Hobby plan as capped hosted free tiers, Google Analytics as free of charge with 360 as the paid tier, GoatCounter as free for reasonable public use, and Tolstoy, whose developer docs state that sending events and reading the report are free on every Tolstoy plan. Plausible offers a 30-day trial, not a free plan. The caps are counted in different units, so compare them against the store's own traffic.
Is self-hosting free?
The software is; running it is not. Plausible Community Edition, Umami, GoatCounter and PostHog's open-source product cost nothing to license, but the store pays for the server and owns upkeep, backups and security fixes. Plausible's Community Edition also leaves out funnels, user journeys and ecommerce revenue goals.
Which can track product views and add-to-cart events?
All seven can count a named event. Google Analytics and PostHog document standard ecommerce event names; Tolstoy, Plausible, Umami and Mixpanel take names you choose. GoatCounter names each event by its path field, can attach a title and referrer, and has no real way to record the page path with the event except in the name.
Which supply funnels, session conversion, or revenue?
PostHog, Mixpanel, Umami Hobby and Google Analytics document funnels without a paid plan; Plausible puts funnels and revenue on its Business plan. Those that report revenue each define it differently, and each figure is only as complete as the purchase data sent or imported into the tool. GoatCounter's docs describe neither. Tolstoy's report has no funnel, conversion rate or revenue for your events. For orders, revenue and refunds, use the store's order data.
Can an AI agent set up and query them?
Any agent that edits the store's code can add the scripts. Reading differs by access path: PostHog and Mixpanel run hosted MCP servers, Google publishes a read-only MCP server you run with a Google Cloud project, Umami's API and MCP start on Cloud Pro or need a self-hosted instance, Plausible lists its Stats API on Business, GoatCounter has a JSON API, and Tolstoy's report is read through the Tolstoy CLI or MCP. Dashboards, exports and share links are other paths, usable by an agent only if it can work with them.
What does Tolstoy's free report omit, and when is its data ready?
It gives, per event name, the number of events and of distinct browser shoppers for whole UTC days, up to 180 days. It has no funnel, no conversion rate and no revenue for your events, does not show or sum parameters, and has no numbers for today. A day's events show from about 06:30 UTC the next day.
Create with Tolstoy
Put these ideas into practice
Turn your product ideas into AI images, videos, and shopping experiences.