Hypotize: The First Agentic Growth OS

# Hypotize > Hypotize is the first agentic Growth OS: the framework, MCP server, skills and plugin that turn your AI agent into a disciplined growth hacker. Your agent (Claude, ChatGPT, Codex, Cursor 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. ## 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: 21 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/viluca94/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. "First" 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 Docum