How to add Poppy to your website

The simplest way to add Poppy to your website is to let Kite do it, because every site Kite builds publishes the poppy.json file AI agents look for, and on any other site you publish that small file yourself at https://yourdomain.com/.well-known/poppy.json with your business name, your domain, and at least one way agents can work with you.

Key Takeaways

  1. Kite adds Poppy to your website for you. Every site Kite builds publishes a Poppy file automatically, a site Kite moves over from another platform gets one too, and on a site whose code lives in your own GitHub repository, Kite adds the file when you ask and your own deployment publishes it.
  2. Adding Poppy means publishing one small JSON file, poppy.json, at https://yourdomain.com/.well-known/poppy.json. Version 1.0 of the protocol requires only three things in it: the protocol version, your business name and domain, and at least one way for agents to work with you, such as your website. A business that only offers its website needs no sign-in server and no new code.
  3. If you aren't on Kite, your website builder decides most of the work. About three in four small businesses build their sites on platforms like Wix, Squarespace, WordPress, or Shopify, according to a 2025 Clutch survey. Wix and Shopify publish a Poppy file for you, Webflow lets you upload one on Premium plans, and Squarespace gives you no way to serve one at the required address today.
  4. What the file lists changes what agents do. In tests by Scelto, Meta's Muse called a store's own tools within minutes once the file listed them on the store's domain, and clicked through pages when the file offered nothing it could use.
  5. To check your site, open the file's address in a browser, then run your domain through a checker such as Kite's free Poppy checker, which reads your public file against version 1.0.
On this page

What does it mean to add Poppy to your website?

Adding Poppy to your website means publishing a small file called poppy.json at one fixed address on your domain, so that AI agents acting for your customers can find out who you are and how to work with you. The simplest way to do it is to have your website platform publish the file for you, and Kite does that on every site it builds. Poppy is the short name for the Personal Agent Protocol, an open standard led by Sierra and Meta. It describes how a person's own AI agent, such as Meta's Muse, deals with a business on that person's behalf: browsing its website, calling its APIs, or talking to its support agent.

The file is the front door. When a personal agent arrives at your site, it requests https://yourdomain.com/.well-known/poppy.json before anything else. Sierra says Muse "automatically looks for Poppy first on every site it visits" (Sierra). The file tells the agent your business name, confirms which domain speaks for you, and lists the ways in that you offer.

For most small businesses, that file is the whole job. Version 1.0, released in mid-November 2026, requires the protocol version, an organization with a name and domain, and at least one channel. A channel is one of three things: your website, an API agents can call, or your own support agent (Personal Agent Protocol specification). Muse's team says "a website-only setup can begin with publishing this file without adding customer sign-in" (Muse). Customer sign-in, APIs, and conversations with your own agent are optional layers you can add later.

The harder part is usually the address, not the file. Hosted website builders control what lives at /.well-known/, so on some platforms the file appears for you, on some you can upload it in a few clicks, and on some you can't place it at all. That's why this guide starts with what Kite does for you, then goes platform by platform for sites built elsewhere.

If you want the background on what Poppy is and why Sierra and Meta built it, read What is Poppy, the Personal Agent Protocol, and does your business need it? first.

How does Kite add Poppy to your website?

Kite publishes a Poppy file on every site it builds, so there is nothing for you to write, upload, or configure. Every site with Kite's built-in content editor, which includes every new site Kite builds, serves the file at /.well-known/poppy.json on the site's own domain. Which route applies to you depends on where your site lives today:

  1. Kite builds your site. The Poppy file is there from the day the site goes live.
  2. Kite moves your site over from another platform. The moved site becomes one of Kite's sites, so it gets the Poppy file too. This is the way to get a Poppy file if your current platform can't serve one, as on Squarespace today.
  3. Your site's code lives in your own GitHub repository. Import the repository into Kite and ask it to add a Poppy file. Kite makes the change the same way it makes any other change to your code, and your own deployment publishes it. If you protect your main branch, Kite opens a pull request instead, and the file goes live when you merge it.

The file Kite publishes does more than say who you are. It names your business and points agents to a description of two things they can do on your site. First, they can read the text of every page in your sitemap in a single file, so an agent answering its person's question about your services gets the whole answer in one request instead of clicking from page to page. Second, they can send you a message, such as a booking or quote request, which reaches you by email. Agents use the site as guests, without signing in to customer accounts, and sending that message is the one action the file offers them.

Kite keeps the file current for you. When you edit a page in the content editor, the text agents read changes with it, and the site's llms.txt describes the same two things, so every AI system that visits gets the same picture of your business. Those two tasks are the ones a small business's customers ask agents to do most, which the section on what your Poppy file should offer agents explains.

To confirm the file is live on your domain, run it through Kite's free Poppy checker, which is described in the section on checking your site below.

Does your website already have a Poppy file?

If your site isn't on Kite, check first whether your platform already publishes a Poppy file for you, because most sites that have one got it from their platform rather than writing it themselves. PAP Checker, an independent checker that keeps a directory, had found 10,815 sites serving a poppy.json on November 20, 2026. Wix made 7,443 of them and Shopify made 3,271. Only 68 sites, belonging to 64 companies, had published their own (PAP Checker).

A sample of ordinary small businesses shows the same pattern. Pico, an AI agent run by Håkon Åmdal, requested the file from the websites of 1,636 small Norwegian companies on November 17, 2026, and 1,450 of them answered. Ninety served a valid file: 67 made by Wix and 23 by Shopify. "None was written by the company," Pico reports (Pico).

You can check your own site in under a minute:

  1. Open https://yourdomain.com/.well-known/poppy.json in your browser, using your real domain.
  2. If you see a short block of text in curly braces that includes protocol_version and your own domain, your site already has a Poppy file.
  3. If you see a "page not found" error, your normal homepage, or a login screen, your site has no Poppy file at that address.

A file your platform made is worth reading even when it works, because it speaks for you. Pico found that the business name in each file is whatever the platform had on file, and 18 of the 67 Wix files carried a placeholder or a run-together web address instead of the company's name, such as "Mysite 1" (Personal Agent Protocol GitHub, issue 91).

Does your website builder support Poppy?

Whether you need to do anything depends mostly on where your site is built. That matters for most small businesses: in Clutch's 2025 survey of 406 US small business owners, 41 percent built their sites with no-code builders like Wix or Squarespace and 34 percent with platforms like WordPress or Shopify (Clutch). This table shows where each common platform stood on November 20, 2026.

Platform

Publishes a Poppy file for you?

What you do

Sites Kite builds or moves over

Yes, for every site with Kite's built-in content editor

Nothing to set up

Wix

Yes, version 1.0, with a tool server agents can call for site search and business details

Check the file and the business name in it

Shopify

Yes, version 0.4, pointing agents to Shopify's shopping protocol

Check the file; you can't edit it

Webflow

No

Upload poppy.json to a "well-known" folder in the Assets panel, on a Premium Site plan or higher

WordPress

No

Serve the file with a plugin that publishes well-known addresses

WooCommerce stores

No

Use a Poppy plugin, and check which protocol version it speaks

Squarespace

No

You can't serve a file at the required address today; moving the site to Kite gives you one

Custom-coded sites

No

Put the file at public/.well-known/poppy.json, or ask Kite to add it to your repository

Wix

Wix publishes a version 1.0 file for the sites it hosts, with no setup from the owner. Each Wix file lists one tool server on the site's own domain, which an agent can use to search the site and read the business details. Pico found that all 67 Wix files in its sample described that server with the same words: "Site search, business details, and site actions such as browsing products and services, booking, and starting a purchase" (Personal Agent Protocol GitHub, issue 91). Scelto, which tested Wix sites on November 15, says Wix also links an older version 0.4 copy for agents that haven't moved to 1.0 (Scelto).

Two limits apply. Sites on a free Wix address such as user.wixsite.com/site get no file at all, because the protocol identifies a business by its domain, not by a folder on a shared one (Personal Agent Protocol GitHub, issue 117). And because Wix writes the file, you can't add entries of your own to it.

Shopify

Shopify publishes a Poppy file for every store. Sierra says merchants "don't need to do any extra implementation" (Sierra). The files declare version 0.4, the release candidate before 1.0, and list one entry that points agents to the store's Universal Commerce Protocol endpoint, Shopify's protocol for carts and checkout (Stacktree). Allbirds' file, read on November 20, 2026, has exactly that shape (Allbirds).

The version number has a practical cost. Before 1.0, the protocol treats versions as compatible only when they match exactly, so an agent that implements only 1.0 must not use a 0.4 file. "Shopify generates the file, so the update has to come from Shopify," notes Pollen, a free open-source checker (Pollen). Scelto adds that merchants can't edit the file or add their own agent to it (Scelto). On Shopify, your job is to open the file once and confirm it names your store and your domain.

Webflow

Webflow lets you upload well-known files yourself. Its help center gives four steps: open the Assets panel, create a folder named "well-known", upload your files to it, and publish the site. When you publish, "Webflow automatically moves all files from the well-known folder to your site's /.well-known/ directory" (Webflow). A file named poppy.json keeps its name, so it lands at /.well-known/poppy.json.

Adding new well-known files requires a Premium Site plan, a legacy Business Site plan, or higher. Files must be .json, .txt, or .md and no larger than 100 KB, and the feature doesn't work on webflow.io staging addresses, so test on your real domain.

WordPress and WooCommerce

On WordPress, a plugin can serve the file for you. The Well-Known URIs plugin adds a settings page under Settings, then Well-Known URIs, where you define an address under /.well-known/, choose its content type, and paste the file's contents, with no code (WordPress.org). For Poppy, the address is poppy.json, the content type is application/json, and the contents are your file. You need administrator rights to use the settings page.

WooCommerce stores have a fuller option. Agent Adapt, a free plugin, publishes /.well-known/poppy.json along with sign-in, cart, and order support, and its makers say there is "nothing to configure" (WordPress.org). Its directory page says it implements draft 0.2 of the protocol, though, and an agent that implements only 1.0 won't use a 0.2 file. Check what version a plugin speaks before you rely on it.

Squarespace

Squarespace doesn't publish a Poppy file, and it doesn't give you a way to serve one at the required address. Squarespace is on Sierra's list of design partners, but PAP Checker's directory listed no Squarespace sites among the platforms publishing files on November 20, 2026. A Squarespace customer reported in the Squarespace forum that page addresses there can't contain a "." character, which rules out a page at /.well-known/poppy.json (Squarespace Forum).

The usual workaround for root files doesn't fit Poppy. Beyondspace, which sells Squarespace plugins, shows how to upload a file and add a URL mapping that redirects your address to the file's location on Squarespace's content delivery network (Beyondspace). That works for some verification files. Under Poppy's rules, though, an agent uses a file only when the final address after every redirect ends in exactly /.well-known/poppy.json, and only when the domain inside the file matches the host that served it. A file sitting on Squarespace's network fails both checks.

Until Squarespace adds support, the way to give a Squarespace site a Poppy file is to move the site. When Kite moves a site over from another platform, the moved site publishes a Poppy file like every other site Kite builds.

Custom-coded sites

If your site is your own code, put the file in your public folder. In Next.js, Vite, or Astro, that means public/.well-known/poppy.json. Pollen warns that "some hosts and static site generators skip folders that start with a dot, so check that the URL actually returns the file" (Pollen). If you've imported the repository into Kite, you can ask Kite to add the file instead, and your own deployment publishes it.

How do you create a poppy.json file?

A complete version 1.0 file for a business that only offers its website fits in five lines. Here is one for a fictional bakery:

{
  "protocol_version": "1.0",
  "organization": { "name": "Corner Street Bakery", "domain": "cornerstreetbakery.com" },
  "web": {}
}

Each part does one job:

  • protocol_version says which version of the protocol the file follows. Use "1.0". From 1.0 on, a newer minor version such as 1.1 won't break agents built for 1.0 (Personal Agent Protocol specification).
  • organization gives your business name and your domain. The domain is your identity. Stacktree's field guide puts it this way: "The name is only a label: the domain is the identity" (Stacktree). You can also add logo_uri, a secure web address for your logo.
  • web is your website. An empty web entry, written as {}, tells agents they may browse your site as ordinary signed-out visitors (Pollen).

Leave out the auth section if customers can't sign in through an agent. That section is optional in 1.0, and leaving it out tells agents you don't offer sign-in. Kite's own file for kite.ai is this shape: the version, the organization with a logo, and an empty web entry (Kite).

If you have something agents can call, list it in apis. Each entry needs a type (openapi for an OpenAPI description or mcp for an MCP tool server), a secure web address, and a short description. The protocol's own guides say a language model usually reads those descriptions to choose what to call, so describe each one in terms of the customer's task and keep it short (Personal Agent Protocol guides). Stacktree's example description reads "Search the catalog and check stock. No account needed." (Stacktree). The sites Kite builds use this part of the file to point agents to the description of their two tasks, reading the site and sending the owner a message.

List only what you have built. Accessibility Checker's guide notes that "every URL in the file must actually work, so list browser_session_endpoint, an API or an agent only once you have built it" (Accessibility Checker).

Be careful with examples you find elsewhere. Many were written against the October 2026 drafts. Stacktree lists fields that are gone in 1.0, including poppy_domains, sign-in types inside poppy.json, and the fixed scopes poppy:read and poppy:write, since companies now name their own scopes (Stacktree). When Kite asked AI answer engines on November 20 how to write the file, their sample files declared versions 1.0, 1.1, and 0.1, and one used key names the specification doesn't have.

Where does the file have to live?

The file has to be served over HTTPS at exactly https://yourdomain.com/.well-known/poppy.json, and the domain written inside it has to match the host that serves it. The specification sets four rules (Personal Agent Protocol specification):

  • The address is fixed. The file must be at /.well-known/poppy.json on your domain or a subdomain you control, over HTTPS.
  • Redirects are allowed, within limits. The address may redirect, for example from www.yourdomain.com to yourdomain.com. Every hop must be HTTPS, and the final address must end in exactly /.well-known/poppy.json with nothing after it.
  • The domain must match. The domain in your file must be the host that finally serves it, after any redirects. Agents compare the two ignoring capital letters but nothing else, so www.yourdomain.com and yourdomain.com count as different.
  • Every address in the file is HTTPS. That covers any API or endpoint you list, though not the ordinary pages of your website.

The reasons are about trust. The exact-path rule exists because otherwise "anyone who can serve a file from some path on a domain, such as an upload, could redirect to it and have it speak for that domain." The domain match "stops a document copied to another domain from speaking for the Company" (Personal Agent Protocol specification).

Three mistakes come up often:

  • Your homepage answers instead of the file. Many sites return their normal page for any address they don't recognize. Kin Lane of API Evangelist checked 24,577 API providers in October and found 2,525 answering the address with something that wasn't a Poppy file at all. "A 200 is not adoption," he writes (API Evangelist).
  • The file sits only on another host. A file that speaks for api.yourdomain.com doesn't speak for yourdomain.com, so agents looking at your main domain find nothing (Stacktree).
  • A firewall or bot check blocks it. If your security settings challenge automated visitors, agents may never see the file. The Agent Adapt plugin's site health check looks for exactly this, a cache or firewall intercepting the discovery documents (WordPress.org).

Do you need a developer to add Poppy?

You don't need a developer to add Poppy to your website. On a site Kite builds, the file is already there, and on a platform that lets you upload files, a website-only file takes a few minutes. You need a developer, or a coding agent, only when you want to offer agents more than your website. The specification says a business without a sign-in server "can offer signed-out Sessions, including APIs, conversations, feedback, and browser join, without running an authorization server" (Personal Agent Protocol specification). Stacktree, which has run Poppy in production since the first draft, says the signed-out start "is one static file and can ship the day you write it" (Stacktree).

The work grows when agents should do more than read your pages. Letting them call your systems means verifying the signed header every agent request carries. Letting them act inside a customer's account means adding sign-in with a consent page. Stacktree calls sign-in the most expensive stage: "We had a first version live in two days on an OAuth server we already ran, and spent the next four weeks correcting it. Do not start here" (Stacktree).

The protocol's authors built that larger job for coding agents. The official company implementation kit is a skill that agents such as Claude Code, Codex, Cursor, Gemini CLI, and GitHub Copilot can follow (Personal Agent Protocol). It works in five steps:

  1. Survey. The agent reads your code and settings and lists the systems you already run.
  2. Design. It maps the protocol onto those systems and writes a plan. You choose what to offer, and "building starts only after you approve the design."
  3. Build. It builds one stage at a time, checking each before moving on.
  4. Verify. It runs the kit's validator, which acts as a test agent, and works through a checklist for things a tool can't test, such as consent wording.
  5. Report. It drafts a report for the protocol's maintainers, and nothing is sent until you file it.

The validator "starts real Sessions and sends test messages to your company agent," so point it at a test copy of your site, not your live customer flows. If you already use a sign-in vendor, it may do part of the work. Clerk's Next.js package serves the file for you when you turn on its poppy option (Clerk), and Auth0 says its upcoming beta serves the file and the matching sign-in metadata (Auth0).

What should your Poppy file offer agents?

Offer agents the channel that answers the questions your customers ask most, on your own domain, because what the file lists decides whether an agent works through it or falls back to clicking through your pages. The protocol's guides tell companies to "start with the tasks you want personal agents to accomplish and the interfaces you already have" (Personal Agent Protocol guides).

Scelto, which builds Poppy support for WooCommerce stores, tested Muse on a test store on November 14 and 15, 2026, and changed one thing at a time (Scelto):

  • With a file that listed tools on Scelto's own separate API host, Muse "tried, but its browser mistyped our host name and wandered off."
  • With a tool server on the store's own domain, Muse "called it within minutes," searching products and then reading one.
  • With a conversation endpoint on the store's domain, Muse opened a real conversation with the store's agent and quoted its shipping answer.
  • With order support added, Muse showed the shopper an approval card, placed a pending order when they said yes, and handed over the store's own payment link.

The same tests show what happens when a file offers nothing Muse can use. On a large Shopify store whose file pointed only to Shopify's shopping protocol, Muse read the file, then clicked through every colorway for about four minutes. By its own account, "No Poppy/UCP call was made." On GoodRx, whose file listed only a sign-in server, Muse browsed page by page. Muse also browsed when a tool's answer was incomplete: once Scelto's store-policy tool returned the full return conditions instead of a summary and a link, Muse answered without opening a page (Personal Agent Protocol GitHub, issue 117).

The version you publish matters too. On November 14, Muse spoke only versions 0.3 and 0.4 and ignored a file that offered only 1.0. Scelto reports that since November 18, Muse sends version 1.0 (Personal Agent Protocol GitHub, issue 117).

For a small business, the useful tasks are usually simple: read what you offer, check your hours and prices, and ask for a booking or a quote. That's why the sites Kite builds list two things in their Poppy file: the full text of the site in one file, and a way to send the owner a booking or quote request. An agent can answer its person's question from the first and act on it with the second, without guessing its way through your contact form.

How do you check whether your website supports Poppy?

A site supports Poppy when https://yourdomain.com/.well-known/poppy.json returns a valid file whose version, name, domain, and channels pass the specification's checks. Opening the address in a browser tells you whether a file exists. A checker tells you whether an agent will accept it.

Several free checkers read your public file:

  • Kite's Poppy checker reads your public discovery file against version 1.0. It follows up to five redirects, checks that the final address is exact, and confirms the version is 1.0 or a later 1.x, the name is present, the domain matches, and at least one channel is listed. Each check shows the evidence it read, a fix, and the section of the specification behind it. It also suggests tasks agents could do on your site, based on your homepage.
  • Pollen is an open-source command-line checker that tests a live domain or a local file before you publish, and prints a pass or fail for each rule.
  • PAP Checker checks a file against whichever version it declares, from 0.1 to 1.0, using only read requests, and lists each problem, warning, and note by the rule it tests (PAP Checker).
  • Stacktree's Poppy Simulator goes further than reading. It acts as a test agent, starts a signed-out session, tries the first API you list, and sends requests a company must refuse (Stacktree).
  • The official validator in the company implementation kit tests everything, including sign-in, but it starts real sessions, so run it against a test copy of your site (Personal Agent Protocol).

Know what a pass covers. Kite's checker says plainly that "a pass covers public discovery only" and isn't an official certification. A pass means agents can find and trust your file, not that every tool behind it works. Also check that the tool you use speaks 1.0. Flow's checker, for example, is labeled for draft 0.1 and still checks for poppy_domains, a field 1.0 removed (Flow).

Poppy is one part of how agents read a site. To see how your whole site looks to AI agents and crawlers, Kite's AI agent readiness check runs the 24 checks a marketing website controls and ranks the fixes by how much each one helps.

Common questions

Is this the same Poppy as Poppy AI or the Poppy popup builder?

No. This guide covers Poppy, the Personal Agent Protocol from Sierra and Meta, which works through a poppy.json file. Other products share the name, including a popup builder and an AI landing-page tool called Poppy AI, and they have nothing to do with a file at /.well-known/. The overlap confuses AI assistants too. When Kite asked ChatGPT "How do I add Poppy to my website?" on November 20, 2026, it asked which Poppy was meant, and Claude didn't mention the protocol at all.

Which Poppy version should I publish?

Publish version 1.0. It's the current version, Kite's checker and Pollen test against it, and Scelto reports that Muse has sent version 1.0 since November 18, 2026. If you also want agents that still speak 0.4 to use your site, the protocol's guides say you can publish the older copy at its own address, such as /poppy/0.4.json, and list it under supported_versions in your main file. Static files are enough (Personal Agent Protocol guides).

Can I edit the Poppy file Shopify or Wix publishes for me?

Not as of November 2026. On hosted commerce platforms only the platform writes the file, so neither you nor an app you installed can add entries to it. Scelto has proposed, in the protocol's public repository, that platforms let merchants approve entries from installed apps, and that proposal is still open (Personal Agent Protocol GitHub, issue 91).

Does a Poppy file replace robots.txt or llms.txt?

No. Robots.txt still tells crawlers what they may fetch, while Poppy tells a visiting agent who you are and how it can work with you (Accessibility Checker). An llms.txt file is a plain-text guide to your site for AI systems, which does a different job from the Poppy file. The sites Kite builds publish both, and the two files describe the same two tasks.

Will agents use my Poppy file once it's live?

Muse checks for the file first on every site it visits, according to Sierra, and Scelto's tests show it using a store's tools when the file lists them on the store's own domain. Other agents are behind. In Scelto's tests, Dot by OpenAI never used poppy.json, and Pico saw requests for the file on its own four domains only from readiness scanners over eight days. The file is cheap to publish and Muse already reads it, so the practical step is to publish it now and add channels as agents start using them.

Sources

Questions this page answers

  • How do I add Poppy to my website?
  • How do I make my website support the Personal Agent Protocol?
  • How do I create a poppy.json file for my website?
  • How do I check whether my website supports Poppy?
  • Does my website builder support Poppy?
  • Do I need a developer to add Poppy to my website?

About the author

Devon Wells leads product marketing at Kite, the AI marketer. He spent a decade marketing industrial AI, from self-driving trucks to computer vision, but now Kite handles most of the work itself.

Tell me your goal and I’ll get to work.

Get started for free