Image: Pomerado
Why it mattersAgent teams have been driving browsers through sites with no API. A reusable MCP tool that repairs itself when a page changes is a different plan.
Most AI agents still work on a website by driving a browser one click at a time. The run is slow. It breaks when the page changes. Pomerado launched on Y Combinator today with a different plan. It turns a website into a Model Context Protocol tool once. Any agent can then call that tool. The company says it is in the YC Fall 2026 batch. Its launch page had two votes two hours after it went live.
The founders are Alan Ma and Akshay Manglik. Pomerado says both came from McKinsey and Scale AI. They built agents for Fortune 500 customers in finance, healthcare and software. Pomerado says the slow part on those builds was always the integration. A legacy portal with no API would take weeks. The company says its own tool can wrap a new site in under 30 minutes from a single prompt.
What the open-source side is
The core minting harness is on GitHub under the Pomerado organisation. The repository is MIT-licensed. Its description reads "YC F26. Turn any website into an API / MCP in minutes." The repo had 148 stars at the time of the launch. The Pomerado GitHub organisation was created on 16 September 2026. It has four public repositories.
A developer can skip the hosted service and run the open-source core locally. The company also publishes a keyless MCP endpoint. An agent can try the hosted version without a sign-up. The install step for a Claude Code or Codex user is one mcp add command against mcp.pomerado.ai.
The numbers the company reports
Pomerado's own claims on the launch page are specific. The company says its deterministic integrations are 2 to 4 times faster than a browser agent. It says they are 10 to 1000 times cheaper. It says they are about 30 percent more reliable. One worked example on the home page is an employee onboarding task across four internal systems. Pomerado finished it in 8 minutes 16 seconds for 3 cents. Codex Computer Use took 21 minutes 22 seconds and cost $41.01. Both figures are the company's.
A second feature the company highlights is self-healing. The example on the site is a healthcare insurance claim lookup. The portal moved its date filter into a new panel. Pomerado's run failed. The company says its maintenance step caught the break, patched the filter step, and reran the call. The agent got its result instead of an error.
The company does not name customers on the launch page. By its own account, the product is in private pilot with teams in finance, healthcare and operations.
Where this fits for a builder
The agent plumbing is splitting into two layers. One is the browser agent for sites nobody will ever wrap. That is Computer Use, browser-use, and the frontier model's own browser mode. The other is a reusable tool a model can call. That is an MCP server, a scoped OAuth integration, or a durable REST wrapper. Pomerado sits in the second layer. Its claim is that an agent team can replace a slow, brittle browser step with a cheap, stable tool call.
The company's asks on the launch page include intros to teams building AI in insurance, healthcare, legal and logistics. That is where its hosted product is most useful today. Those are regulated workflows that live on legacy supplier portals. A browser agent still cannot handle them reliably at scale.
Source
Pomerado launch on Y Combinator, with the open-source core at Pomerado/pomerado on GitHub and the product at pomerado.ai.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.