# Hypotize

> Hypotize is the agent-native Growth OS: the framework, MCP server, skills and plugin that turn your AI agent into a disciplined growth hacker. Your agent (Claude, ChatGPT, Grok Bot, Cursor, Codex or any MCP client) reads and writes the experiment backlog, scores ideas, moves experiments through their stages and records outcomes and learnings, under rules the server enforces, so the team keeps one queryable history of what was tested and what it taught.

Hypotize works before and after an experiment runs. It does not assign traffic, serve feature flags or run A/B tests in application code: teams keep running tests in their own analytics and experimentation stack.

Who uses it: growth, product, marketing and CRO (conversion rate optimization) teams, and the agencies that run experiments for clients. For CRO, Hypotize holds the test backlog, the ICE or PIE score of each idea, the stage of every test and the archive of results and learnings; the A/B testing tool still runs the test.

## What the Growth OS is made of

- Framework: hypothesis-first ideas, ICE, PIE, RICE or a custom scoring framework, enforced stages, and a measured result with a written learning at the end of every test.
- MCP server: the tools the agent uses to read and update the record.
- Skills: `hypotize-growth-experiments` (the method), `hypotize-weekly-review` (a read-only pipeline review) and `hypotize-backlog-import` (moving a spreadsheet or Notion backlog in batches of 50).
- Plugin: one install for Claude Code and Codex with the MCP server and the skills. Source: https://github.com/hypotizeai/hypotize-plugin. Install guide: https://hypotize.ai/skills
- App: the board, backlog and learnings the team works in, on the same record as the agent.
- Rules: an experiment cannot skip a stage and cannot close without a measured result and a written learning, whether a person or an agent is writing. The agent never touches billing or members.

"Agent-native" means built from the start for an outside AI agent, with no AI of its own, and it covers the experiment programme around the tests. Experimentation platforms such as GrowthBook, Statsig and Optimizely offer MCP servers to set up and read the A/B tests they run; Hypotize works alongside them.

## Connect an AI agent

- MCP endpoint: https://api.hypotize.ai/mcp (Streamable HTTP, OAuth 2.1 with PKCE, dynamic client registration and Client ID Metadata Documents)
- [MCP guide](https://hypotize.ai/mcp): how to connect Claude, ChatGPT, Grok Bot, Cursor, Codex and other clients
- Codex: `codex mcp add hypotize --url https://api.hypotize.ai/mcp`, then `codex mcp login hypotize`
- Scopes: `hypotize:read`, `hypotize:write`, `hypotize:delete` (delete stays off unless the user turns it on at consent)

## Automations and code

- [REST API](https://hypotize.ai/developers): read projects, experiments and learnings, add ideas, move tests through the pipeline and record outcomes from n8n, Make, Zapier, Pipedream or your own code. Base URL https://api.hypotize.ai/v1, personal API key as Bearer, Idempotency-Key on writes.
- OpenAPI 3.1: https://api.hypotize.ai/v1/openapi.json

## What the agent can do

- Projects: list, get context and writing language, create, update, delete
- Experiments: list the backlog and pipeline, get one with votes, action plan and comments, create an idea, update, import up to 50 existing experiments at once, delete
- Scoring: ICE, PIE, RICE or a custom framework; vote and recalculate priority
- Lifecycle: move an experiment to its next stage, record the outcome, set the action plan, add comments
- Learnings: list outcomes and learnings across the project
- Profile: set the user's own name and, optionally, role, only as the user gives them
- Every write tool takes an idempotency key, and project roles are checked on every call. Hypotize makes no server-side model calls.

## Shared workspaces

Every project belongs to a workspace with its own plan and members. New accounts start with a personal workspace and can join others by email invitation. Workspace members can create projects and edit every shared project. Private projects are available on Team and are visible only to explicitly invited people and workspace owners and admins. The highest workspace or project role applies.

Workspace owners and admins manage members and all projects, including private projects. Only the workspace owner controls billing, workspace deletion and ownership transfer. Project guests keep their own roles: contributors can read, vote and comment; viewers can read. Neither counts as a billable member.

`hypotize_projects_list` lists accessible projects across all workspaces and includes `workspace_id` and `workspace_name`. `hypotize_project_create` accepts an optional `workspace_id`: without it, creation uses the user's only workspace membership, or their personal workspace when they have several. The same tools and idempotency keys apply.

## Pricing

- Free: $0, 1 member, 1 project per workspace, unlimited free contributors and viewers.
- Solo: $19.90/month per workspace, 1 member, unlimited projects, unlimited free contributors and viewers.
- Team: $39/month for up to 3 members, $10/month per extra member. Unlimited projects, private projects, and unlimited free contributors and viewers.
- Members are workspace owners, admins and members plus editor or owner project guests, counted once per person. Pending paid invitations reserve members. AI agents and MCP connections never count.
- Workspace overview, unlimited MCP clients and complete experiment history are available on every plan. Monthly billing only; no annual billing.

## Guides

- [Experiment memory](https://hypotize.ai/guides/experiment-memory)
- [Hypotize and A/B testing tools](https://hypotize.ai/guides/ab-testing-tools): Hypotize does not run or measure A/B tests. It keeps the backlog, stages, outcomes and learnings around them, across every tool and channel. The page lists, per A/B testing tool, its documented AI agent access and archive of past tests.
- [The best growth experiment management tools in 2026](https://hypotize.ai/guides/growth-experiment-management-tools): Our pick for teams whose AI agent runs their growth experiments is Hypotize: built for an outside agent, rules enforced on what the agent writes, no AI of its own to pay for. Covers Hypotize, Growth Method, Efestra, GrowthLab and GrowthOrange, plus Notion, Airtable and spreadsheets, each with the team it suits best, its documented AI agent access, its price and what to watch out for. Also lists the learnings archives inside A/B testing platforms, the closed TrueNorth, and renamed or moved products (GrowthHackers Experiments former URL, UpGrow, Effective Experiments). Prices and features reviewed 15 to 22 October 2026; product status checked 28 October 2026.
- [The best MCP servers for marketing and growth teams (2026)](https://hypotize.ai/guides/mcp-servers-for-marketing): Official MCP servers only, grouped by job: experiment records (our pick: Hypotize, whose server holds the agent to the experiment method; Growth Method for campaigns), analytics (Google Analytics, Mixpanel, Microsoft Clarity, Contentsquare), ad platforms (Meta Ads, Google Ads, TikTok Ads, Amazon Ads, Microsoft Advertising, Supermetrics), SEO (Ahrefs, Semrush, DataForSEO, Similarweb), CRM and email (HubSpot, Salesforce, Klaviyo, Customer.io, Intercom, Brevo), design (Canva, Figma, Webflow) and automation (Zapier, Make, n8n). Each of these entries says where the server runs, whether it reads or writes, and what to watch out for. Also covers the MCP servers of A/B testing platforms, Notion, Airtable and Google Sheets, and the tools where we found no official server: LinkedIn Ads, Reddit Ads, Google Search Console, Mailchimp Marketing. Vendor docs reviewed 28 October 2026; A/B testing platforms 16 October 2026; Notion, Airtable and Sheets 22 October 2026.
- [ICE framework](https://hypotize.ai/frameworks/ice)
- [PIE framework](https://hypotize.ai/frameworks/pie): PIE was built to rank pages: Potential, Importance and Ease from 1 to 10, averaged, to decide where to test first. Free ranker for up to 10 pages, with CSV download.
- [RICE framework](https://hypotize.ai/frameworks/rice)
- [Growth experiment template](https://hypotize.ai/templates/growth-experiment-template): free CSV and one-page doc, no account or email. Each field says when it becomes required (to create, to reach Ready, to close). Column names match the Hypotize import, so an AI agent can import the file.
- [Hypotize vs Spreadsheet: Google Sheets and Excel](https://hypotize.ai/compare/hypotize-vs-spreadsheet)

## Who it is for

- [Hypotize for agencies](https://hypotize.ai/for/agencies): one project per client in one workspace, client guests as free viewers or contributors, private projects on Team, and a workspace overview that compares every account.
- [Hypotize for CRO teams](https://hypotize.ai/for/cro-teams): a scored backlog of test ideas, pre-registration the server enforces, variants and the winning variant, evidence linked by test ID, and an archive of learnings the agent reads first.

## Comparisons

- [Hypotize vs experiment trackers](https://hypotize.ai/compare): Growth Method documents an MCP server next to its own AI agents (reviewed 19 October 2026); Notion and Airtable document workspace MCP servers with no stage rules (reviewed 22 October 2026); GrowthOrange, Efestra, and GrowthLab document no access for an outside AI agent (reviewed 15 October 2026).
- [Hypotize vs GrowthOrange](https://hypotize.ai/compare/hypotize-vs-growthorange)
- [Hypotize vs GrowthLab](https://hypotize.ai/compare/hypotize-vs-growthlab)
- [Hypotize vs Growth Method](https://hypotize.ai/compare/hypotize-vs-growth-method)
- [Hypotize vs Efestra](https://hypotize.ai/compare/hypotize-vs-efestra)
- [Hypotize vs Notion](https://hypotize.ai/compare/hypotize-vs-notion): a Notion database as an experiment tracker. Notion documents a hosted MCP server, so an agent can read and write the database; it describes no required properties or stage rules (reviewed 22 October 2026).
- [Hypotize vs Airtable](https://hypotize.ai/compare/hypotize-vs-airtable): an Airtable base as an experiment tracker. Airtable documents an official MCP server on all plans and automations for reminders; required fields are documented only for forms (reviewed 22 October 2026).

## Legal and support

- [Privacy policy](https://hypotize.ai/privacy)
- [Terms of service](https://hypotize.ai/terms)
- Support: hello@hypotize.ai