Protocols & news

What is PAP? The Personal Agent Protocol from Meta and Sierra explained

Listen · 10 min

Narrated by an AI voice. Code examples are on the page.

Short answer

PAP, the Personal Agent Protocol (also called Poppy), is an open standard from Meta and Sierra that defines how a shopper's personal AI agent finds a business, says who it is and who it acts for, and gets access, as a guest or signed in. It does not handle payment itself: for shopping and checkout it hands off to UCP. Version 1.0 was published on 14 November 2026.

Key takeaways

  • PAP covers how a personal agent identifies itself and gets access.
  • It has no payment layer; checkout goes through UCP.
  • Version 1.0 replaced the drafts in November 2026.

What is PAP?

PAP, the Personal Agent Protocol, is an open standard for how a person's own AI agent works with a business on that person's behalf. Its authors call it Poppy for short. It covers how the agent finds the business, proves which agent it is, starts a session for its user, and, if the person agrees, signs in to their account.

The specification puts it this way: Poppy "defines how a Personal Agent, acting for one User, works with a Company on that User's behalf." A personal agent here means an assistant that acts for one shopper, as opposed to a crawler that reads pages for an index.

The problem it targets is familiar to store owners. The protocol site describes today's choice as "all or nothing": either a business blocks the agent, or the agent signs in as the person and can do anything they can. PAP adds steps in between, so a store can let any agent browse, let a signed-in agent see orders, and ask for the person's approval before larger actions.

For how PAP fits with the other standards, see our guide to agentic commerce protocols.

Who is behind PAP?

Meta and Sierra are developing it. Sierra announced the protocol on 6 October 2026 as "an open standard Meta and Sierra are developing along with industry partners at Genesys, Instinct, Rocket, Shopify, Stripe, and Walmart that defines how personal agents interact with businesses."

The project has grown quickly since then, according to Sierra:

  • 9 October 2026: Sierra published a first draft and named 35 more design partners, including Bank of America, OpenAI, Visa, United Airlines, Target and Cloudflare.
  • 13 November 2026: Sierra said version 1.0 was available, with more than 100 design partners in total. (The protocol's changelog dates the 1.0 release 14 November.) It said Shopify had enabled Poppy for every merchant on its platform, and that Poppy is on for everyone who uses Muse, Meta's AI agent.

These adoption claims come from Sierra's announcement. Design partners take part in the design process; being on the list does not mean a company has shipped PAP. The protocol website says it is licensed under the Apache License 2.0.

How does a personal agent get a guest session on your store?

The agent reads a file on your domain, then signs each request it sends to your APIs and your company agent. In version 1.0, a guest visit does not need a login server at all.

1. Discovery. The store publishes a JSON file at a fixed address:

GET /.well-known/poppy.json HTTP/1.1
Host: shop.example.com

It names the organization and lists what agents can use: the website, APIs such as an MCP server or an OpenAPI description, and the company's own agent, if it has one.

2. Agent identity. The agent's ID is an HTTPS address that serves a small metadata document, including a link to its public keys. A store can fetch that document to check which agent it is dealing with.

3. The session header. On each request to your APIs or your company agent, the agent sends a signed token in a Poppy-Session header. It names the session and an ID for the shopper. That ID is stable, so you can recognise a returning visitor, but it is different at every company and must not be derived from personal information such as an email address. The specification says plainly: "Signed-out interactions need no OAuth server."

The spec also says agents should sign these requests with Web Bot Auth, so existing bot defences can recognise the operator before the request reaches your application. Our guide to signed agent requests explains how that works.

Earlier drafts worked differently. In draft 0.2, published 24 October 2026, a guest agent exchanged a signed assertion at the store's token endpoint for a short-lived session token bound to the agent's key with DPoP. Version 0.3, published 4 November 2026, removed that token exchange for signed-out sessions. If you read about PAP guest sessions before November, check which version the description follows.

What happens when the shopper signs in?

Sign-in is standard OAuth. The person signs in on the store's own page and approves the access the agent asked for, such as reading orders or starting a return.

To find where to sign in, the agent reads two standard metadata files:

  • Protected resource metadata (RFC 9728), at /.well-known/oauth-protected-resource on your domain, which names your login server.
  • Authorization server metadata (RFC 8414), published by that login server, which lists its token and revocation endpoints and must list your domain.

The agent then gets an access token for the scopes the person approved. A store may bind that token to the agent's key with DPoP, so a copied token is useless without the key. The spec says personal agents should support DPoP, and leaves the choice to bind to the company. The person or the store can revoke access later.

A store that only wants guest visits can skip all of this. The spec allows a company with no login server, as long as it offers no sign-in.

What does PAP not do?

PAP does not move money. The core specification defines no payment method, and the project roadmap lists payments as future work, with the goal of using "existing payment and checkout protocols rather than replacing them."

For shopping, PAP hands off to UCP. The UCP extension, which Sierra says Shopify co-authored, splits the work: PAP "finds the Company, identifies the Personal Agent and a pseudonymous User, and signs the User in", and UCP handles carts, checkout and payment. Sierra's 1.0 post sums it up: "UCP handles the shopping itself, like finding products, filling a cart, and checking out."

PAP also does not decide what an agent may do on your store. It gives you a standard way to say it. You still choose which agents to allow, which channels to offer and which account scopes agents can request.

How does PAP relate to MCP, UCP, ACP, AP2 and A2A?

PAP sits in front of the other protocols. It answers who is calling and for whom, then points the agent to the channel that does the work.

Protocol Job Relationship to PAP
MCP Lets agents call tools and read data One of the API types a poppy.json can list; signed-in calls carry a PAP access token
UCP Catalog, cart, checkout and orders Linked through PAP's UCP extension; UCP does the shopping
ACP Agent checkout from OpenAI and Stripe Not mentioned in the PAP spec
AP2 Signed proof of what a person approved for payment Not mentioned in the PAP spec
A2A Agent-to-agent discovery and messaging Not mentioned; PAP defines its own conversation format for talking to a company's agent
Web Bot Auth Signed requests that identify the operator Recommended alongside PAP's session header

Where a cell says "not mentioned", the PAP documents we read do not describe a link. That does not rule one out later.

What should a store owner do now?

Start by checking whether you already have it. Sierra says every Shopify store now publishes a Poppy file automatically, so Shopify merchants may not need to do anything. Look for /.well-known/poppy.json on your domain.

If you are not on Shopify:

  1. Get the basics right first. Agents still need clean product data, an llms.txt file and, ideally, an MCP server.
  2. Decide what guests may do. List the tasks any agent can handle, such as stock checks or return policies, and the ones that need a signed-in shopper.
  3. Build against 1.0. The changelog says: "Version 1.0 is published. Build against 1.0." Anything built on the drafts may need changes.
  4. Keep checkout on UCP. If you want agents to buy, PAP alone is not enough.

Note. PAP is new. Version 1.0 arrived in November 2026, and Sierra's adoption figures had not been independently confirmed at the time of writing. Check the changelog before you build.

How Sill helps. Sill supports PAP 1.0 guest sessions on Cloudflare Worker and WordPress plugin installs. It publishes your site's poppy.json and checks the signed session header a personal agent sends with every request to your MCP server, refusing any request that fails the check. Shoppers cannot sign in to their accounts through it yet, and since PAP carries no payment, orders still go through your rules, which approve, decline or hold each one for a person, with every decision signed and recorded. Run a free readiness scan to see where your site stands.

Questions people ask

Why is PAP also called Poppy?

Poppy is the protocol's own short name. The specification says Personal Agent Protocol is "Poppy for short", and the discovery file is named poppy.json.

Is PAP open source?

The protocol website says it is licensed under the Apache License 2.0, and Sierra says it is open for anyone to implement.

Do I need PAP to sell to AI agents?

No. Agents can still read your pages, use an MCP server or check out through UCP without it. PAP adds a standard way for a personal agent to identify itself and the person it acts for.

Can a personal agent see my customers' personal details through PAP?

Not as a guest. Until the person signs in, the specification gives the company only an opaque user ID that must not be derived from personal information. Account data needs a sign-in the person approves, limited to the scopes you grant.

Sources

  1. Sierra: Introducing Personal Agent Protocol · checked 16 Nov 2026
  2. Sierra: Sharing a draft of Personal Agent Protocol · checked 16 Nov 2026
  3. Sierra: Releasing Personal Agent Protocol 1.0 · checked 16 Nov 2026
  4. Personal Agent Protocol · checked 16 Nov 2026
  5. Personal Agent Protocol: Specification (version 1.0) · checked 16 Nov 2026
  6. Personal Agent Protocol: Specification (version 0.2) · checked 16 Nov 2026
  7. Personal Agent Protocol: Changelog · checked 16 Nov 2026
  8. Personal Agent Protocol: Introduction and roadmap · checked 16 Nov 2026
  9. Personal Agent Protocol: UCP extension · checked 16 Nov 2026

Written by The Sill Signals Team. Last reviewed 17 Nov 2026.

Share this articleLinkedInXBlueskyEmail

Keep reading

See which AI agents visit your site, and decide what they may do.

Scan your site free