Turn websites into tested APIs.
Give Peel a URL and say what data you need. It builds a typed API, tests it on the live site and hosts it for your code over REST and your AI agents over MCP.
Free plan: 10 builds/mo and 3,000 runs/mo. The live call needs no account.
- Explore the sitenews.ycombinator.comRunningDone
- Write the API2 endpoints, typed fieldsRunningDone
- Test it on the live siteexample calls passRunningDone
- Checked by a second AIapprovedRunningDone
- Publishv1, ready to callRunningDone
Endpoints appear once the build passes its tests.
- list_stories
- title, url, points, author
- get_story
- title, text, comments
- 1A website goes inYou give Peel a URL and say what data you need.
- 2Peel builds and testsIt finds where the site gets its data, writes the API and tries it on the live site.
- 3Your API comes outCall it from code over REST, or from AI agents like Claude Code and Cursor over MCP.
Faster than a browser agent
Agents click through pages on every call. Peel's model works once, at build time. Every call after runs the same tested program, with no model.
Scraper, agent or API?Cleaner than a scraper
No HTML to parse and no selectors to fix. Each endpoint returns named fields with types.
Peel vs FirecrawlYours to keep
Call it hosted, or export the source and run it anywhere, without a Peel account.
Peel vs PomeradoTry one now. No account needed.
Peel built this Hacker News API and keeps it live. Press Run to call it from this page, paste the call into a terminal, or add the keyless MCP server to your agent. Each address gets 25 free calls a day, and all calls without an account share 2,000 a day.
Call it
curl -X POST https://api.peel.sh/v1/marketplace/kd74n2353r515czbkq7kaet9tx8jctwy/endpoints/front_page_stories:call \
-H "Content-Type: application/json" -d '{}'claude mcp add --transport http peel-try https://api.peel.sh/mcp/keyless
Get typed JSON back
{
"status": "success",
"data": {
"stories": [
{
"title": "Java4Shell: RCE in OpenJDK when processing Base64",
"link": "https://calif.io/java4shell/",
"points": 620,
"author": "werrett"
},
… 29 more
]
},
"execution_time": 0.413,
"transport": "http_replay"
}Request access for more calls.
REST, MCP and OpenAPI.
Every endpoint works on all three, and they all run the same tested version.
Call it from code
curl -X POST https://api.peel.sh/v1/apis/<api-id>/endpoints/list_stories:call \
-H "X-API-Key: $PEEL_API_KEY" \
-H "Content-Type: application/json" \
-d '{"limit": 10}'{
"tool": "call_endpoint",
"arguments": {
"api_id": "<api-id>",
"endpoint_name": "list_stories",
"params": {
"limit": 10
}
}
}curl https://api.peel.sh/v1/apis/<api-id>/exports/openapi \ -H "X-API-Key: $PEEL_API_KEY"
Connect your agent
{
"mcpServers": {
"peel": {
"url": "https://api.peel.sh/mcp"
}
}
}Claude Code · Cursor · Codex · Claude Desktop · any MCP client
Tested before it goes live.
Peel publishes your API only after it has worked on the real site.
Tried on the real site
- Peel calls your new API on the live site, with the exact code it will publish.
- The answer has to come from the site itself, not from a saved copy.
- A second AI, separate from the one that built it, checks the result and can stop it from going live.
- 1.Java4Shell: RCE in OpenJDK when processing Base64620 points by werrett
- 2.ID.me terms required for IRS sign-in177 points by DeepLogin
- 3.Google's AI answers reducing search traffic by 40%163 points by slashdev
The same 30 stories as the page, read from the site in 0.41 seconds.
Nothing is sent until you approve.
- Operation
- create_issue
- Request
- POST /issues.json
- subject
- Checkout page times out
- priority
- High
After you approve
Nothing has been sent yet.
You approve.Peel sends it once and saves the result.
Your code or AI agent sends the same approval again.Peel sends nothing and returns the saved result.
The site never answers.Peel doesn't resend. It asks you to check the site first.
Actions wait for your OK
Your API can also change things on a site, like opening a ticket or placing an order. Peel shows you exactly what it will send, and sends nothing until you approve.
Some actions can't be undone, so try them somewhere safe first.
Your passwords stay out of the AI
Save a site's password and 2FA once. When your API needs to sign in, Peel types them in. The AI that builds your API never sees them.
If a site texts or emails you a code, Peel asks you for it then.
Blanks, never the values.
Own your integration
Export any version as source code and run it without a Peel account.
curl https://api.peel.sh/v1/apis/<api-id>/exports/source \ -H "X-API-Key: $PEEL_API_KEY"
What you can ask for
Describe the data in plain words. Peel names the endpoints and types every field.
list_storiesget_storysearch_productslist_repositoriesget_repositoryPricing
Free to start. Pro adds more builds, runs and parallel builds. Full pricing and run rules
Business
Custom
- Custom builds and runs
- Build concurrency by agreement
- Dedicated support
- Custom integrations
- Build, revision or rebuild: 1 build
- Runs: standard 1, signed-in or premium-network 2, browser and action 3
- Failed runs and builds are free, except builds that time out
- Automatic repairs are free
- Build attempts per day, failed ones included: Free 20, Pro 120, Business 500
Frequently asked questions
Anything else? Email kaan@peel.sh.
What does Peel do?
Give Peel a URL and say what you need from the site. It investigates the site and compiles named endpoints, each with input and output schemas, examples, linked evidence and diagnostics. You call them over REST or MCP, or export them as OpenAPI.
How is this different from a browser agent or a scraping API?
Peel uses models only while it builds. A saved call runs the compiled program over HTTP or a guarded browser with no model involved, so every call follows the same compiled logic. A site can still change or return something unexpected, and the diagnostics say so when it does. Instead of a page to parse, you get named fields that match a schema.
Which websites work?
Public pages are the core case: documentation, directories, listings, pricing pages, news and similar sites. How well a build goes depends on how the site is made. Some sites will fail to build, and Peel refuses some targets under its policies. A failed build keeps diagnostics that show where it stopped.
Can Peel work with pages behind a login?
In preview. Save a site's login on the Logins page, including an authenticator secret, or pass the values with a call. Codes sent by text or email are asked for when they arrive. The API asks for each value by name, and Peel fills it in only on the exact site they were given for. The values, and the one-time codes made from them, stay out of model prompts, generated code and saved programs. Builds that sign in have passed end to end on nine self-hosted business apps so far, and other sites are unproven. The signed-in endpoints guide in the docs has the current list.
Can Peel submit forms or take other actions?
Yes, through action endpoints. A write runs only after you confirm the exact operation, its input and the requests it will send. Peel records that confirmation before sending, so confirming twice returns the first result instead of sending again. If Peel cannot tell whether a write went through, it marks the outcome unknown and never resends it on its own. Some actions cannot be undone, so try them somewhere safe first.
What happens when a site changes?
Peel re-runs each API's saved example inputs every day and checks each result against the endpoint's output schema. When the site or the program fails twice in a row, it rebuilds the API with the same instructions, at most once a week, and the new version replaces the current one only if every endpoint keeps its name and exactly the same inputs, outputs, auth and pagination. Repairs do not count against your builds. You can turn repair off for any API, rebuild it yourself, or roll back to an earlier version from its revision history. There is no guaranteed time to fix a broken endpoint.
How do plans, builds and runs work?
Every plan includes monthly builds and runs: Free ($0/mo) 10 builds and 3,000 runs, Pro ($50/mo) 100 builds and 30,000 runs. Business volume is agreed with you. A build is one new API, revision or rebuild. A run is one call to a published endpoint that returns a result: a plain HTTP run counts 1, a signed-in or premium-network run counts 2, and browser and action runs count 3. Failed runs and failed builds do not count, except a build that runs out of its time budget. Run packs add prepaid runs that do not expire, with optional auto recharge. Settings shows your current use.
How do I call an endpoint?
Over REST, send a POST to the endpoint's call URL with your API key. Over MCP, add https://api.peel.sh/mcp to your client and your endpoints show up as tools. Both run the same published version, which you can also export as OpenAPI or as a source folder that runs without a Peel account.
What data does Peel keep, and who can see it?
Peel keeps your account and billing records, what you submit for builds, what builds produce (fetched content, request traces, schemas and versions) and usage logs. Other users cannot see your APIs unless you publish one to the marketplace. Model providers process build content while Peel builds, but private sign-in values are kept out of their prompts. Peel does not sell personal information. You can delete your account in Settings or by email, and the Privacy Policy has the details.
Are there rate limits?
Yes. API calls, MCP requests and builds are rate limited, and the limits rise with your plan. Free runs 1 build at a time, Pro 4 and Business up to 12. A request over the limit gets HTTP 429 with a Retry-After header. Abusive load can be throttled or blocked under the Acceptable Use Policy.
Is it legal to use Peel on a website I don't own?
That depends on the site, the data and the law where you are. You need the right to access each site you point Peel at, and you must follow its terms and the law that applies to you, including privacy law for personal data. A successful build does not mean a use is allowed. The Terms and the Acceptable Use Policy set out your responsibilities.
I run a website and see traffic from Peel. What can I do?
Email kaan@peel.sh with your domain, the URLs and times involved, and what you want done. A person reviews every report and can disable specific endpoints, take down marketplace listings or block builds and calls against your domain. The Acceptable Use Policy lists what to include.
How do I get support or report a security issue?
Email kaan@peel.sh for support, billing, privacy and security reports. Each plan's support level is listed under Pricing, and the security contact is also published at /.well-known/security.txt. Peel does not hold a SOC 2 report or other compliance certification today.
Build your first API.
Paste a URL, describe the data, and call the result from your code or your agent.
Free plan: 10 builds/mo and 3,000 runs/mo.