Pomerado, a YC F26 Startup, Sells Websites as MCP Tools

Pomerado, a Y Combinator F26 startup, charges $50 a month to turn a website with no API into an MCP tool an AI agent can call, and its free tier stops at 100 tool calls a day. The company launched on Launch HN, where the post had 42 points and 35 comments as of Nov. 11. MCP, the Model Context Protocol, is a standard that lets an agent call external tools.

The pitch, as the founders wrote it, is that agents are limited by their ability to use legacy systems, which often lack APIs and may ban computer use. Pomerado's agent explores a site in a browser and completes a task once, then generalises that run into a repeatable API. A second model the company calls Guardian reviews the calls against guardrails, and synthetic-input tests check for overfitting. A maintenance job repairs the tool when runs fail, for example after a button is renamed. The post says calls to a finished tool usually return in 5 to 30 seconds.

What the $50 tier buys

Pomerado's pricing page lists three plans. The page names no investors beyond "Backed by Y Combinator" and gives no funding amount, no team names and no customer names. The logos it shows, including Delta, Costco, Kayak and OpenTable, appear as integration examples, not as confirmed customers. The Launch HN post cites unnamed users: a personal trainer, car dealerships and builders of personal assistants.

Plan	Price	Tool calls a day	New integrations a month
Free	$0	100	10
Premium	$50 a month	1,000	100
Business	Contact sales	100,000 or more	1,000 or more

Which parts are open source

The core is on GitHub under the MIT licence, copyright Pomerado AI, Inc. Versions 0.1.2 and earlier were under AGPL-3.0-only, a licence that obliges network services to publish their source, so the relicensing to MIT loosens the terms for anyone building a product on it. The repository had 1.6k stars, 136 forks, 21 open issues and 49 open pull requests when fetched, and its releases section was empty.

The README's setup needs macOS or Linux, Node 24.21 or later, an OpenAI API key used for building integrations, and an MCP client. It warns that a run "does not check authority or intent", that a mint can perform a write action once during the build, and that cancelling a job does not undo actions already sent to a website. It describes bot-protection handling and breakage repair only as features of the paid Pomerado Cloud. The self-healing maintenance that the launch post leads with is therefore not in the MIT code.

What commenters pressed on

Commenters raised three issues, and the founder's answers are the company's own account.

• On reliability, a commenter said a tool can return HTTP 200 while making no useful progress. The founder reported a roughly 2.2 percent run error rate, with 82 percent of those errors resolved automatically. No outside party has checked either figure.
• On bot defences, a commenter doubted that Amazon would work. The founder said it is one of the most-used tools and runs on a custom stealth stack with CAPTCHA solvers.
• On legal exposure, commenters cited sites that ban automated access. The founder said Pomerado does not recommend using it on LinkedIn.

On why OpenAI or Anthropic would not build this natively, the founder pointed to anti-bot infrastructure and token-margin concerns. A commenter asked for a maintenance dashboard; the founder said there is none, because repairs happen at execution time.

The benchmark is the company's own

The pricing page claims Pomerado is "2x faster, 10 to 1,000 times cheaper, and 30% more accurate" than several browser-use tools, based on 30 tasks across 25 business apps, each run once. A sample onboarding task is shown finishing in 8 minutes 16 seconds for 3 cents, against 21 minutes 22 seconds and $41.01 for Codex Computer Use. By that example the cost gap is about 1,370 times, outside the page's own range, and one run per task cannot show how often either tool fails.

The Terminal has covered other agent tooling, including Vercel's agent-browser and Atuin Threads. Neither source says whether Pomerado has raised money. The next observable events are a disclosed round or a named paying customer, and a tagged release in the repository.