News & Insights

Poppy Protocol: What Sierra's Personal Agent Protocol Means for Brands

Sierra released Poppy, the Personal Agent Protocol, as v1.0 on November 13, 2026, with every Shopify store publishing a Poppy file and Meta's Muse reading it. Here is how it works, how it sits next to ACP, UCP, MCP and A2A, and what consumer brands should do while adoption is early.

Sakshi Gupta

Table of contents

Talk to us

Summarize using AI

ChatGPT Custom Icon
Claude Custom Icon
Gemini Custom Icon
Perplexity Custom Icon
Grok Custom Icon

On November 13, 2026, Sierra released version 1.0 of the Poppy protocol, short for Personal Agent Protocol, an open standard for how a person's AI agent identifies itself and works with a company, according to Sierra's release post by co-founders Bret Taylor and Clay Bavor. The same post says every Shopify store now publishes a Poppy file automatically and that Meta's Muse agent looks for Poppy first on every site it visits.

For a consumer brand, Poppy changes how a shopper's agent arrives at your store, not what it finds there. It gives the agent a standard way to say who it is, ask permission and learn what it may do. On Shopify, the actual shopping still runs through Google's Universal Commerce Protocol (UCP), so your catalog, policies and landing pages remain the parts that decide the sale.

What is the Poppy protocol?

Poppy, short for Personal Agent Protocol, is an open standard for how a person's AI agent identifies itself, signs in and works with a company. The v1.0 specification names three parties: the User, the Personal Agent acting for that user, and the Company the agent visits. A company can serve the agent through its website, its APIs or its own customer service agent, which the spec calls the Company Agent.

Sierra, which builds customer service agents for large companies, first announced the effort on October 6, arguing that agents still load pages and fill forms the way people do. A first draft followed on October 9. Version 1.0 reflects what design partners learned over five weeks of testing against real websites, APIs and agents. The spec is published under the Apache License 2.0, and its GitHub repository has been public since October 14.

How does the Personal Agent Protocol work?

A company publishes a discovery file, the agent reads it to learn which channels and permissions exist, and the shopper approves account access through OAuth. It builds mostly on existing standards:

  1. Discovery file. A company publishes poppy.json at /.well-known/poppy.json on its domain. It lists the organization, sign-in details and scopes, conversation protocols, APIs, a web browsing option, a tasks endpoint and extensions. The agent must check that the domain in the file matches the site that served it.

  2. Agent identity. Every request carries a signed Poppy-Session header that identifies the agent, so a company can tell which agent is visiting even when the shopper is signed out.

  3. Sign-in and scopes. When a task needs the shopper's account, the agent uses standard OAuth 2.0, either a direct sign-in or a device sign-in. Each scope carries hints saying whether it makes reversible or irreversible changes and whether it touches personal data. The company should show a consent page, and the shopper can disconnect the agent later from account settings.

  4. Channels. APIs can be described as OpenAPI documents or remote MCP servers. Conversations with the Company Agent mark every message as written by an agent or a human and support handoff to a person.

  5. Tasks and feedback. A company can list tasks it does not support through the protocol, point to alternatives, publish skills for agents to follow, and collect reports when an agent could not finish a task.

  6. Extensions. Industries can add features without changing the core. Sierra says Shopify co-authored the commerce extension, published as the UCP extension.

Adoption can be staged. Sierra's implementation kit says a company that offers only a website can stop after publishing poppy.json without sign-in, then add APIs, conversations and account access later.

Who backs Poppy, and who is live?

More than 100 companies are design partners, but a handful are live: Shopify publishes Poppy files, Meta's Muse reads them, and three partners have launched.

  • Initial partners (October 6). Meta, Genesys, Instinct, NiCE, Rocket, Shopify, Stripe and Walmart, per the release post. Shopify's Mani Fazeli said the protocol helps define how agents and merchants handle product questions, purchases, returns and exchanges.

  • Draft partners (October 9). 35 more, including OpenAI, Visa, Mastercard, PayPal, Adyen, Klaviyo, Gap Inc., Nordstrom and Target, per Sierra's draft post.

  • v1.0 partners (November 13). 60 more, including Salesforce (Commerce Cloud and Agentforce), Wix, Wayfair, HelloFresh, Braze, Checkout.com and Affirm.

  • Live today. Mercury, Synchrony and Turkish Airlines are going live, Shopify stores publish the file, and Muse is the only consumer agent Sierra names as reading it.

Google and Amazon do not appear in Sierra's published partner lists. OpenAI is a design partner, but neither company has said ChatGPT reads poppy.json. For background on the agent doing most of the early work, see our explainer on Meta Muse.

How is Poppy different from ACP, UCP, MCP and A2A?

Poppy handles the arrival, who the agent is and what it may do, while ACP and UCP handle buying and MCP and A2A are plumbing.

Protocol

Started by

What it covers

How it relates to Poppy

Status for brands, November 2026

Poppy (Personal Agent Protocol)

Sierra, October 2026, with 100+ design partners

Agent identity, sign-in, permissions, channels and feedback, across any industry

The protocol itself

v1.0 released November 13; auto-published on Shopify; read by Muse

UCP (Universal Commerce Protocol)

Google, January 2026, co-developed with Shopify, Etsy, Wayfair, Target and Walmart

Catalog, cart, checkout, orders, identity linking and product questions (Ask), with a profile at /.well-known/ucp

Poppy's UCP extension makes UCP the checkout and payment mechanism

Live on Shopify stores; powers checkout in Google AI Mode and Gemini

ACP (Agentic Commerce Protocol)

OpenAI and Stripe, September 2025

Checkout between an agent and a merchant, with payment passed as a secure token and the merchant of record unchanged

Not referenced in the Poppy spec; no ACP extension exists

Open standard; Shopify says ChatGPT purchases complete on the store's own checkout

MCP (Model Context Protocol)

Anthropic, November 2024; now under the Linux Foundation's Agentic AI Foundation

How an agent calls tools and data on a server

One of the two API types Poppy supports

Shopify's UCP endpoints run over MCP; also used by Shop MCP

A2A (Agent2Agent)

Google, April 2025; now under the Linux Foundation

Messages and task handoffs between agents, with a payments extension (AP2)

Not referenced; Poppy defines its own conversation format

Supported by more than 150 organizations, mostly enterprise use; a listed UCP transport

For a brand, these are not competing choices; your commerce platform picks them. For the commerce layer in more depth, see our guides to ACP, UCP and the difference between ACP and UCP.

What does Poppy look like on a Shopify store today?

On Shopify, the Poppy file is generated automatically and points agents to the store's UCP profile, so checkout runs through UCP rather than Poppy itself. We fetched poppy.json from several Shopify stores on November 18, and the files followed the same template:

  • One conversation channel, of type ucp, pointing to the store's /.well-known/ucp profile. There were no OpenAPI, MCP or web browsing entries in the Poppy file itself.

  • One scope, for preparing checkouts as a signed-in customer, flagged as irreversible and touching personal data. Its description tells agents that most stores allow guest checkout, so sign-in is usually unnecessary.

  • The UCP extension, version 1, which under the extension rules makes UCP checkout the approval point: the shopper reviews terms and the agent completes the purchase once. When a step needs the shopper's own browser, the checkout returns a continue_url.

  • A declared protocol_version of 0.4, the release candidate, not 1.0.

The linked UCP profiles declared version 2026-08-25 and served shopping and checkout over MCP, the same machinery behind Shopify Shop MCP. Merchants cannot edit the generated file. A GitHub issue opened November 13 notes that neither the merchant nor an installed app can add entries, so a support app that handles returns or order status cannot be listed for agents to find.

In plain terms, a Shopify brand is Poppy-discoverable today without doing anything; what the agent does next depends on UCP and your product data.

What should a consumer brand do about Poppy now?

Don't build anything custom yet; confirm your file exists, make sure your product data and policies can stand alone, and watch which agents actually arrive. In order:

  1. Check your domain. Open /.well-known/poppy.json and /.well-known/ucp on your storefront domain. On Shopify both should exist. On other platforms, ask your vendor; Salesforce Commerce Cloud and Wix are design partners, but neither has announced a ship date in Sierra's posts.

  2. Hold off on a custom build unless you run your own stack. Brands with in-house commerce or account systems can use Sierra's kit, starting with the file alone. Most brands gain little from going further until more agents than Muse read it.

  3. Make the catalog complete. Through the UCP extension, the agent's real work is catalog search and checkout, so categories, named variant options, barcodes and spec-rich descriptions decide whether you are picked. Our guide to catalog enrichment for AI shopping agents covers the fields.

  4. Write policies an agent can quote. Returns, shipping, warranty and sizing answers should be explicit text, because UCP's Ask capability answers product questions from your own data.

  5. Plan for the hand-off. Escalated checkouts and product links send the shopper to your site mid-task. That page should continue what the agent promised; see our guide to funnels that convert AI agent traffic.

  6. Log agent visits. Poppy gives each agent a signed identity. Ask your platform or CDN whether that shows up in logs, so you can see which agents visit and what they do.

  7. Watch the roadmap, not the headlines. Payments outside UCP, push notifications such as shipping updates, and finer permissions are still listed as future work.

What hasn't Sierra said about Poppy yet?

Sierra has not named a governing body, published adoption numbers, or settled how payments work outside UCP, and Shopify's files still declare draft version 0.4. The open questions:

  • Governance. The spec is Apache 2.0 on GitHub, but unlike MCP and A2A, which sit under the Linux Foundation, no neutral foundation runs it. Sierra also sells the business-side agents Poppy connects to.

  • Adoption. Sierra has published no count of live sites, agent sessions or orders. Three live partners plus Shopify and Muse is the public record.

  • Payments. Sierra's October post listed payment extensions as future work. Today, checkout and payment for stores run through UCP.

  • Which agents read it. Muse is the only named consumer agent. ChatGPT, Gemini, Claude and Copilot have not said they read poppy.json.

  • Version gap. Under the spec's own compatibility rules, pre-1.0 versions only match the same minor version, so Shopify's 0.4 files and a strict 1.0 agent are, on paper, different versions until Shopify updates.

  • Merchant control. Shopify has not documented an admin setting for its Poppy file, and the proposal to let apps add entries is still open.

  • ACP and A2A. The spec does not mention either, so how a ChatGPT checkout or an A2A agent fits alongside Poppy is undefined.

How Nudge helps brands get ready for Poppy

Poppy standardizes how an agent introduces itself; what it finds afterward is still your catalog, your policies and your pages. Nudge Catalog Enrichment works on the data agents read through UCP, Nudge Funnels on the pages agents and shoppers land on, and Nudge Shopper Insights on whether any of it shows up in AI answers.

Nudge Catalog Enrichment generating Product schema (JSON-LD) for a product page

What the agent or shopper needs

How Nudge handles it

Complete, filterable product data for UCP catalog search and checkout

Catalog Enrichment analyzes the catalog, assigns Shopify taxonomy categories and category metafields, and keeps Google Merchant Center and ACP and UCP feeds consistent, with your team approving changes

Clear answers to product and policy questions

Enrichment workflows write specs, use cases and policy details from your own product facts

A product page an agent can read accurately

Agent funnels serve a machine-legible version of the page to AI shopping agents, fact-identical to what humans see

A landing that continues the task after a hand-off

Organic funnels adapt the existing product URL in place for AI-referred visitors, tested against a live control and shipped only with your approval

Knowing which agents mention you and send shoppers

Shopper Insights tracks prompts, citations and agent analytics across AI platforms

Nudge Funnels building a different page for each arrival: AI platforms, ads and email

Poppy is early, and for most consumer brands it arrives through platforms like Shopify rather than custom builds. The work that pays off whichever protocol wins is the same: product data an agent can trust and a page that finishes the job. For the wider context, see our primer on agentic commerce.

See how Nudge turns clicks into conversions. Book a demo

Frequently asked questions

What is the Poppy protocol?

Poppy, short for Personal Agent Protocol, is an open standard from Sierra that defines how a person's AI agent identifies itself to a company, signs in with the person's permission, and uses the company's website, APIs or support agent. Sierra released version 1.0 on November 13, 2026, under the Apache License 2.0, after five weeks of design partner testing.

Do Shopify merchants need to do anything to support Poppy?

Not to be discoverable. Sierra says every Shopify store now publishes a Poppy file automatically. In files we checked on November 18, 2026, the file pointed agents to the store's UCP profile for shopping and checkout. Merchants cannot edit the generated file today, so the useful work is improving product data, policies and landing pages.

How is Poppy different from UCP?

Poppy covers what happens when an agent arrives: who it is, which channels exist and what it is allowed to do. UCP, the Universal Commerce Protocol Google launched with Shopify and major retailers, covers the shopping itself, including catalog search, cart, checkout and orders. Poppy's UCP extension makes UCP the checkout and payment mechanism for stores that support it.

Does Poppy replace MCP or A2A?

No. Poppy builds on MCP, letting a company describe its APIs as remote MCP servers or OpenAPI documents, and on OAuth for sign-in. It does not mention A2A and defines its own conversation format between a personal agent and a company's agent. MCP and A2A are both governed under the Linux Foundation, while Poppy has no named governing body.

Which AI agents use Poppy today?

As of mid-November 2026, Meta's Muse is the only consumer agent Sierra names as reading Poppy; Sierra says Muse looks for Poppy first on every site it visits. OpenAI is a design partner, but neither company has said ChatGPT reads Poppy files. Google and Amazon do not appear in Sierra's published partner lists.

You don’t control where discovery happens.

You do control whether you show up.