GitHub - Pomerado/pomerado: YC F26. Turn any website into an API / MCP in minutes. Open source core integration minting harness. MIT.

Open original

Pomerado is an open-source tool that builds deterministic MCP integrations for websites, including ones with no API, from a plain-language task description. The core is MIT-licensed and runs locally, while Pomerado Cloud adds hosting, managed logins and breakage repair.

  • Integrations are built from a natural language description alone, with no HAR file or recording needed, and they run deterministically, which the project says makes them faster and more reliable than computer-use agents.
  • Each mint has a read or write authority, and write mints perform the action once during the build; a run only signs in on the site and its declared sign-in origins.
  • It works behind logins and two-factor auth, but the build needs an OpenAI API key, while running a finished integration makes no model calls.

Pomerado is an open-source tool that turns a website into an MCP integration your AI agent can call. You describe the task in plain language, name the site, and say whether the integration should only read or also change things. Pomerado then drives a real Chromium browser, explores the site, completes an example, generalizes it, and saves the result as a folder of deterministic source code. The pitch is that these integrations are faster and more reliable than having an agent operate a browser through computer use, and that they work on sites with no API, including ones behind a login and two-factor authentication. Building one needs an OpenAI API key, Node 24.21 or later, and macOS or Linux, and the server runs under clients like Claude Code, Codex, Cursor, VS Code, Claude Desktop, and Gemini CLI. The part worth paying attention to is authority. Each integration is either read or write, and the agent has to ask you before choosing write. A write mint performs the action once while building, then reviews the source without repeating it. A few limits matter in practice: a run does not re-check authority or intent, edits to generated source aren't reviewed, and sign-in is only replayed against the site and sign-in origins you approved. Pomerado refuses to type your password into a form that submits to a different domain, but it doesn't verify where a page's own scripts send what you enter. If a call fails with possible_commit set, the action may already have happened, so you should check the site before retrying. The integration itself stores no credentials; it asks for your login on each call. Operationally, a mint gets 20 minutes of active work, with waiting for your answers not counting against that budget, and cancelling a job closes the browser but can't undo an action already sent to the site. Each finished integration is a folder with generated operation modules, a pomerado.json schema, a deployment.json holding its tool description and sign-in origins, and a README with add commands for each client. Integrations run on the Pomerado installation that built them, so the README recommends a global npm install for a stable path. The core, including the minter, the review component called Guardian, and the standalone host, is MIT-licensed and needs no account. That makes it a reasonable thing to evaluate yourself, though the real test is how it handles a messy production site rather than the example in the docs.

Ask about this article