Hypotize wordmark
Get started

Field guide · A/B testing tools

Your A/B testing tool measures. Hypotize remembers.

Optimizely, VWO, GrowthBook, Amplitude and most other A/B testing tools can now be read by an AI agent, and many keep an archive of past tests. Each archive covers the tests that ran inside that tool. Hypotize keeps one experiment memory across every tool and channel, including ads, email and pricing tests, and makes your agent follow the method before a test is closed.

Short answer

Hypotize does not run A/B tests. It sits around them.

Hypotize is an experiment tracker built for AI agents. It does not split traffic, serve variants or compute statistical significance, and it has no direct integration with any A/B testing tool. Your A/B testing tool does that work. Hypotize keeps what comes before and after: what to test next, the stage each experiment is in, and every outcome with its evidence and learning. An AI agent connected to both reads the result in the A/B testing tool and records it in Hypotize.

Two different jobs

Measuring a test and remembering a programme

Your A/B testing tool

  • Splits traffic and serves the variants.
  • Computes the result and its statistical confidence.
  • Keeps the history of the experiments it ran, often with a learnings archive.
  • In most cases, an AI agent can read it through an official MCP server.

Hypotize

  • Keeps the backlog before the test: hypothesis, priority score, owner.
  • Tracks every experiment through its stages, wherever it runs.
  • Stores the outcome, the evidence and the learning after the test.
  • Refuses to close an experiment without a result and a learning, for people and agents alike.

Where each memory stops

Every tool remembers its own tests

A learnings archive inside an A/B testing tool is useful, and several are good. Its limit is scope: it holds the experiments that ran in that tool. A growth team also tests ad creatives, email subject lines, offers and prices, often in four or five different systems. Those results end up in decks, chats and spreadsheets, and the next person or agent cannot find them. Hypotize is the one record that covers all of them, whichever tool ran the test.

Tool by tool

14 A/B testing tools, checked on 16 October 2026

Each line comes from the vendor's own documentation, linked in the last column. "Not documented" and "not found" mean we could not find it in the official pages, not that it does not exist.

Optimizely

AI agent access
Official hosted MCP server for all Experimentation customers, Web and Feature Experimentation. It requires Optimizely Agent Platform.
Archive of past tests
Collaboration: hypotheses, briefs with a section for learnings, and saved views used as a results repository.
Where Hypotize fits
Optimizely runs web and feature experiments. Ad and email tests usually run elsewhere. Hypotize keeps one record across all of them, and the agent reads Optimizely results through its MCP server.

Sources: Optimizely MCP, Collaboration

VWO (Wingify)

AI agent access
Official Wingify MCP server for campaigns, reports, heatmaps and recordings. The Claude connector requires an Enterprise plan or the Developer Plus add-on.
Archive of past tests
Wingify Plan: observations, hypotheses and past results. Plan is not among the tools the MCP server lists.
Where Hypotize fits
The agent reads VWO results through the MCP server and keeps hypotheses and learnings in Hypotize, next to tests from other channels.

Sources: Wingify MCP, Claude connector, Wingify Plan

AB Tasty (Wingify)

AI agent access
No MCP server found in AB Tasty's public documentation. AI features (Evi) work inside the product.
Archive of past tests
Learnings Library for campaign results, insights and next actions. Ideas Backlog, in beta.
Where Hypotize fits
Hypotize gives an outside AI agent the backlog and the learnings through a remote MCP server, across AB Tasty and other channels.

Sources: Learnings, Ideas Backlog, Evi

Convert

AI agent access
Official MCP server, hosted and local, for experiences, goals, audiences and reports.
Archive of past tests
Compass: observations, hypotheses and a knowledge base. The official MCP package can read and write them.
Where Hypotize fits
An agent can already work on Convert tests end to end. Hypotize adds the tests that do not run in Convert, such as ads, email and pricing, to the same record.

Sources: Convert MCP, MCP package, Compass

Kameleoon

AI agent access
Official MCP server, including experiment results with significance and the winning variant.
Archive of past tests
Learnings tab: a summary of each experiment's results, generated by Kai. A repository across experiments is not documented.
Where Hypotize fits
Hypotize keeps the memory across experiments and channels, with the learning the team decided to keep.

Sources: Kameleoon MCP, Learnings tab

GrowthBook

AI agent access
Official MCP server, hosted and local.
Archive of past tests
Learnings, since 27 August 2026: a knowledge base for people and AI agents, available through the API and MCP, including qualitative research.
Where Hypotize fits
GrowthBook can analyze experiments run elsewhere when their data is in your warehouse. Hypotize covers the tests whose data is not there, such as ad creatives and email, under the same stage and closure rules.

Sources: GrowthBook MCP, Learnings, Importing experiments

Statsig

AI agent access
Official hosted MCP server for gates, experiments and metrics.
Archive of past tests
Knowledge Bank: a searchable repository of experiment learnings across teams.
Where Hypotize fits
Hypotize keeps the backlog and the tests that run outside Statsig, such as ads and email, in one record.

Sources: Statsig MCP, Knowledge Bank

PostHog

AI agent access
Free hosted MCP server with experiment tools: results, timeseries and lifecycle.
Archive of past tests
No dedicated archive documented. The docs suggest writing conclusions in the experiment's description field.
Where Hypotize fits
Hypotize gives hypothesis, outcome and learning a structured place, and the agent reads PostHog results through its MCP server.

Sources: PostHog MCP, Experiment lifecycle

LaunchDarkly

AI agent access
Hosted MCP server for feature flags, AgentControl, observability and metrics. Experimentation is not listed. It is not available in federal or EU environments, where a local server is the alternative.
Archive of past tests
No learnings repository documented.
Where Hypotize fits
Hypotize keeps hypotheses, outcomes and learnings readable by the agent through its own MCP server.

Sources: LaunchDarkly MCP, Hosted MCP

Datadog Experiments (Eppo)

AI agent access
Datadog MCP server, experiments toolset for results and diagnostics. It is not enabled by default.
Archive of past tests
Eppo's docs describe a Knowledge Base of concluded experiments.
Where Hypotize fits
Hypotize covers the ad, email and pricing tests that do not run in Datadog Experiments.

Sources: Datadog MCP tools, Eppo Knowledge Base

Amplitude Experiment

AI agent access
Official MCP server for every customer, with experiment tools that include variant performance and significance.
Archive of past tests
No dedicated archive documented. The Web Experiment docs say to capture “the goal, key learning, decision made, and implementation follow-up in an internal knowledge base.”
Where Hypotize fits
Hypotize can be that knowledge base, read and written by the same agent that reads Amplitude's MCP server.

Sources: Amplitude MCP, After the experiment

Intelligems

AI agent access
Official MCP server for experiments, analytics and the Shopify catalog.
Archive of past tests
No learnings repository documented.
Where Hypotize fits
Intelligems tests run on Shopify. Hypotize adds the tests on ads, email and other channels outside the store.

Sources: Intelligems MCP

Shoplift

AI agent access
No MCP server found in Shoplift's docs. A public API returns test results with significance, and the docs show how to use it with Claude.
Archive of past tests
No learnings repository documented.
Where Hypotize fits
Hypotize gives the agent the backlog and the history through MCP, while Shoplift results reach it through the API.

Sources: Test results API, API with Claude

Adobe Target

AI agent access
Official MCP server in public beta, with 62 tools that include activity reports.
Archive of past tests
No learnings repository documented.
Where Hypotize fits
Hypotize sits next to it as one experiment memory across every channel.

Sources: Adobe Target MCP

With a connected AI agent

How the two work together

1

Before the test

The agent reads the Hypotize backlog and the learnings of past tests, on every channel, before it proposes the next one.

2

During the test

The test runs in your A/B testing tool, your ad platform or your email tool. Hypotize keeps the stage, the owner and the dated readings.

3

Reading the result

The agent reads the result where it was measured, for example through the A/B tool's own MCP server or API.

4

Closing the loop

The agent records the outcome, the winning variant, the evidence link and the learning in Hypotize. Without a result and a learning, the server keeps the experiment open.

The more A/B testing tools expose an MCP server, the easier this gets: the agent reads the numbers at the source instead of from a screenshot, and Hypotize keeps them next to the hypothesis that started the test.

When you do not need Hypotize

One tool, one channel, one archive

If every test you run happens inside one A/B testing tool, and that tool keeps a learnings archive your agent can read, that archive may be enough. Hypotize earns its place when tests run in more than one system: a web experiment tool, an ad platform, an email tool, a pricing change shipped by hand. It also helps when you want the same rules for people and agents, so nothing is marked done without a result and a learning.

Questions

Hypotize and A/B testing tools

Is Hypotize an A/B testing tool?

No. Hypotize does not split traffic, serve variants or compute statistical significance. Your A/B testing tool runs and measures the test. Hypotize keeps the backlog before it, the stage while it runs, and the outcome, evidence and learning after it.

Does Hypotize integrate with Optimizely, VWO or GrowthBook?

Not directly. An AI agent connected to both does the work: it reads the result in the A/B testing tool, through that tool's MCP server or API, and records it in Hypotize with the evidence link and the learning.

My A/B testing tool already has a learnings archive. Why add Hypotize?

Because that archive covers the tests that ran inside that tool. Ad, email, offer and pricing tests usually run elsewhere. If every test you run happens in one tool, its archive may be enough. Hypotize is for teams whose tests run in more than one system.

Can an AI agent close a test in Hypotize without a result?

No. An experiment moves one stage at a time, and the server refuses to close it without a measured result and a written learning. The agent has to get the result first, for example from the A/B testing tool's MCP server.

Next step

Keep one memory for every test you run

Open the web appExperiment memory guideConnected AI: tools and limits
Comparing experiment trackers instead?

Sources and verification: official vendor documentation linked in the table, checked on 16 October 2026. We did not open vendor accounts or trials, so features behind a login are described only as documented. Vendors change fast: if a line here is out of date, write to hello@hypotize.ai and we will correct it.