Trawl Beta Launch, Plus MCP Upgrades

Updated August 2026. This post announced the private beta in June 2026. Since then the gate came down: Trawl is open, no seat request needed, and the free plan now starts you with 3,000 compute credits rather than 500. One other thing aged badly, and it is called out in place below: tool names are now prefixed-only, so the un-prefixed aliases this post promised would keep working have since been removed. The release notes are otherwise left as they were written.

Give Your Agent a Web Tool That Doesn't Get Blocked

Your AI agent is only as smart as the pages it can actually read, and most of the web now answers a bot with a block screen or a CAPTCHA, not its content. Trawl is a hosted Model Context Protocol server that hands your agent a stealth-tier web-read tool: point it at almost any URL, get structured JSON back, with no browser infrastructure to run.

Your agent shouldn't have to lose a fight with a CAPTCHA to read a web page. Free to start, no credit card.

What You'll Learn

• What Trawl gives your agent
• Latest update: in-app docs and reliability
• What's new in this release
• Reliability and polish
• How to get started
• Key Takeaways
• FAQ

What Trawl Gives Your Agent

You connect Trawl once, and your agent gets a web-read tool built for the sites that fight back, not just the easy ones:

• Gets past anti-bot defenses: stealth-tier rendering handles the bot-detection and CAPTCHAs that stop ordinary scrapers at the door, so your agent reads the page instead of a block screen.
• No infrastructure to run: no headless browsers, no proxy pool, no patching every time a site ships new defenses. Trawl runs all of it; your agent just calls a tool.
• Self-healing extraction: when a site changes its layout, Trawl's AI repairs the extraction instead of returning empty fields, so scheduled jobs keep delivering.
• Structured results: every run returns clean JSON your agent can act on, plus a screenshot and full run history.

Point your agent at a product page sitting behind anti-bot, ask for the fields you want, and you get structured data back. A scraper that returns data instead of excuses, that's the whole pitch.

Want this for your agent? Start free on Trawl. No credit card, 3,000 compute credits at signup.

Latest Update: In-App Docs and Reliability at Scale

Since the beta opened we shipped a larger release (June 2026) focused on two things people kept asking for: documentation you can actually use inside the product, and scraping that holds up when a lot of jobs fire at once.

This release touches the web app, the platform, billing, and security; the MCP server and CLI are unchanged this cycle.

• 🌐 Web app · In-app documentation: a new docs hub now lives inside the app, with role-based guides for three kinds of users, an API reference, copy-with-my-key snippets, and full-text search. The path from "what is this" to "first scrape running" no longer leaves the product.
• ⚙️ Platform · Reliability at scale: scheduled and async runs now go through a job queue with a global concurrency cap, so a burst of jobs landing at once no longer stampedes the browser layer. And when an account hits its limit you see it directly in the web app, a quota-pause badge and an exhaustion banner on the scraps list, plus cron-pause visibility, instead of jobs failing quietly. The reasoning behind building this on the database we already run, rather than a message broker, is its own story.
• 💳 Billing · Fairer metering: premium-proxy bandwidth is now metered to what a run actually uses, and the free-tier copy reflects the one-shot trial honestly.
• 🔒 Security · Hardening: we closed a handful of server-side-request and tenant-isolation issues as part of routine hardening. Nothing you have to do, just the unglamorous work of keeping a multi-tenant scraper safe.

What's New in This Release

Most of this release went into the MCP server, the surface your agent actually talks to. If you are building agents, four changes earn their keep:

• 🔌 MCP · Markdown responses: ask for markdown and your agent reads a clean, ready-to-use answer instead of burning tokens parsing raw JSON.
• 🔌 MCP · Keyset-cursor pagination on list_scraps: listing stays fast and consistent no matter how many scraps you have, so an agent can walk the full set without timeouts.
• 🔌 MCP · Prefixed tool names: every tool now also registers as trawl_*, so it is unambiguous when your agent loads a dozen servers. No longer accurate: at the time this shipped the un-prefixed names still worked. They have since been removed. Only the trawl_-prefixed form is registered today, and there are no aliases. See the current tool list.
• 🔌 MCP · Structured output via outputSchema and structuredContent: typed, predictable results your agent can rely on instead of guessing the shape.

Same endpoint: point your MCP client at api.trawl.me/api/mcp over Streamable HTTP.

Reliability and Polish

The rest of the release is the quiet work that makes the product feel solid:

• ⚙️ Platform · Correct screenshot formats: run screenshots are stored with the right image type, detected from the file itself, so previews always render.
• ⚙️ Platform · Steadier scheduling: a boot smoke-test gate and a cross-pod heartbeat catch cron problems before they reach you.
• 🌐 Web app · A clearer wizard: a brand-glyph catalog, a net-sweep loader, and a centered chat bubble make scrap setup easier to follow.
• 🌐 Web app · Redesigned settings: the scrap settings drawer moved to a master-detail layout, with several rounds of polish on top.
• 🌐 Web app · Better defaults: digest emails are now opt-in, and the chat keeps the previous run's screenshots so you can compare two runs at a glance.
• 🌐 Web app · Safer content: legal pages run through DOMPurify sanitization, and the running app version is visible in the footer.

How to Get Started

This section replaces the original "How to request access". The private beta is over and there is no waiting list.

Sign up at trawl.me and the free plan starts you with 3,000 compute credits, no credit card, which is enough to run real scrapes and watch your agent work before you commit to anything.

Prefer the terminal? npm install -g @trawlme/cli then trawl login gets you a free token in one step.

Key Takeaways

1. Trawl gives your agent a stealth-tier web-read tool that gets past anti-bot defenses, with no browser infrastructure to run.
2. The June 2026 release adds an in-app docs hub and a job-queue-backed scraping engine that stays reliable under concurrency.
3. Self-healing extraction keeps scheduled jobs delivering when sites change their layout.
4. This release sharpened the MCP server: markdown answers, keyset pagination, trawl_* tool names, and structured output.
5. Trawl is open: free plan, 3,000 compute credits, no credit card, no seat request.

If you want managed, stealth-tier web data for your agents without running browser infrastructure, Trawl can help. Try it free.

FAQ

How do I get started with Trawl?

Sign up at trawl.me. There is no seat request and no waiting list: the private beta this post announced has ended. The free plan starts you with 3,000 compute credits and no credit card.

Can Trawl scrape sites that block bots?

Yes, that is the point. Trawl uses stealth-tier rendering designed to get past the anti-bot defenses and CAPTCHAs that stop ordinary scrapers, with no setup on your side. It is not a guarantee for every site, but it handles cases that block most tools.

Does Trawl cost anything to try?

No. The free plan starts you with 3,000 compute credits at signup, with no credit card. That is enough to run real scrapes and evaluate Trawl before choosing a paid plan.

What is the Trawl MCP server?

It is a hosted Model Context Protocol endpoint that lets an AI agent use Trawl as a set of tools: create a scrap, scrape a URL, schedule recurring scrapes, trigger runs, and read structured results. It speaks JSON-RPC 2.0 over Streamable HTTP at api.trawl.me/api/mcp.

Did the MCP tool names change? Will my agent break?

They did change, and this is the one thing in this post that has since reversed. When this release shipped, every tool gained a trawl_ prefix while the original un-prefixed names kept working. Those aliases have since been removed: only the prefixed form is registered today. If your agent still calls an un-prefixed name, update it.

What is keyset-cursor pagination on list_scraps?

Instead of paging by offset, list_scraps returns a cursor that points to the next page. This keeps listing fast and consistent even when you have a large or growing number of scraps.

Which MCP clients work with Trawl?

Any JSON-RPC 2.0 client that supports Streamable HTTP, including Claude Desktop, Cursor, and n8n. Add the api.trawl.me/api/mcp endpoint with an Authorization bearer token and your agent gets the Trawl tools.

Is there a CLI or API as well?

Yes. The @trawlme/cli npm package lets you log in with trawl login, and the full tool and endpoint reference lives at api.trawl.me/api/docs.

Written by Pierre | June 2026 · updated August 2026