Personal agents don’t have great options today.
They can use your credentials and act like you when browsing websites designed for humans (and do a poor job of it), with all of your permissions.
Or businesses prevent them from interacting at all.
Either way, businesses burn cycles figuring out if requests are from agents, and agents burn cycles trying to access websites designed for humans. And you don’t have granular control over what actions your personal agent can take on your behalf.
So we’re excited to announce that Runlayer is a design partner of Personal Agent Protocol (aka Poppy), alongside Stripe, Meta, OpenAI, and a bunch of other great companies.
Personal Agent Protocol, announced in October by Sierra and Meta, provides an open standard for personal agents to interact with businesses on behalf of individuals. Sierra released version 1.0 of the spec on Nov. 13, and the protocol now has 100+ design partners.
The protocol builds on top of common standards businesses are likely already using (OAuth, MCP, OpenAPI) and standardizes the entire interaction between personal agents and businesses, including:
- Discovery: how does the agent find the business?
- Identification: how does the agent prove whose agent it is?
- Authorization: what is the agent allowed to do?
- Task execution: how does the agent perform tasks for the user?
We run into this dynamic a lot at Runlayer, as employees want to give agents (like Muse) access to their company email, calendar, and documents while security teams are left deciding between allowing or blocking access to everything. We think Personal Agent Protocol can help solve this challenge at the layer where individuals’ personal agents interact with businesses.
How does Personal Agent Protocol (aka Poppy) work?
At a high level, businesses expose a file at /.well-known/poppy.json that indicates who they are, what resources they expose to agents (APIs via OpenAPI descriptions and remote MCP servers, their website, an agent representing the business, etc.), and the permissions agents can request to act on behalf of end users.
Users sign in to businesses using OAuth directly on the business’s website to grant permissions to agents acting on their behalf. With Direct Sign-In, the sign-in option the protocol prefers, agents never see user credentials, since it’s a direct connection between the end user and the business. Agents use a single session to make API calls, browse the website, chat with the business’s agent, etc., and signing in adds the access the user granted. Agents sign each request with a short-lived key and a user ID that’s unique per business (so users’ identities aren’t revealed unless they choose to sign in).
Actions can be gated on additional permissions and confirmation using extensions (e.g., an agent might be able to see flight options but need the user to confirm the details before it changes a flight). Businesses can also explicitly list actions agents can’t perform, so agents don’t waste time trying different approaches and can let users know right away.
An example from the Personal Agent Protocol site: let’s say you’re interacting with a hotel’s website. Any agent can look up available rooms, agents acting on behalf of users who’ve signed in can access reservation details, and agents can make reservations on behalf of users with their approval. Businesses can still choose to not allow agents or grant them carte blanche, leaving it up to the business to determine what level of access agents should have to perform different actions.
What’s new in Personal Agent Protocol version 1.0?
After five weeks of design partners building against draft versions of the spec with real-world agents, websites, and APIs, here are some of the highlights from Sierra’s announcement of Personal Agent Protocol 1.0:
- Shopify enabled Poppy for all merchants on its platform
- Mercury, Synchrony, and Turkish Airlines are going live with Poppy
- Meta’s personal agent, Muse, now checks for Poppy first on every website it visits
- 60 new design partners
There’s also an implementation kit including a validator to help businesses implement the 1.0 spec themselves.
Why did Runlayer become a Personal Agent Protocol design partner?
With Personal Agent Protocol, businesses can control what agents are allowed to do, what permissions they need to perform different actions, and when human-in-the-loop confirmation should be required. And users have control over what actions agents can perform on their behalf with secure, fine-grained permissions.
We’ve built these capabilities into Runlayer so the agents employees use at work have:
- On-behalf-of access: employees authenticate as themselves, so agents only have access to what they do
- Fine-grained permissions: control what agents are allowed to do, down to specific tool calls
- Human-in-the-loop: require human approval for risky or destructive actions
We also log every tool call for complete visibility and can export audit logs to your SIEM. Check out our post on agent identity and least privilege to learn why we think agents should have their own identities and only have access to what they need.
What does this mean for businesses?
There are two categories of agents that will soon be accessing your business, namely agents your customers use to interact with your business and agents your employees use to get work done.
Personal Agent Protocol provides a way to control what your customers’ agents can do on your website and with your APIs. If you’re interested in learning more, start by creating a poppy.json file. The implementation kit is a great resource for next steps. If you already expose a remote MCP server, that’s one of the API types Poppy supports, making it a great starting point.
Runlayer provides employees the golden path to use agents at work, ensuring the secure way is the easy way. Gain AI enablement, security, and control all in one platform that works with 20+ AI clients and provides access to 200+ MCPs in the catalog.
Personal Agent Protocol FAQ
What is Personal Agent Protocol?
Personal Agent Protocol (Poppy for short) is an open protocol that standardizes how personal AI agents discover businesses, authenticate on behalf of their users, and complete tasks using websites, APIs, and business agents. Businesses control what agents can do, and users control what their personal agent can do on their behalf.
Who developed Personal Agent Protocol?
Meta and Sierra announced Personal Agent Protocol in October 2026. Over 100 businesses, including Runlayer, Shopify, Visa, Stripe, and OpenAI, are design partners helping shape what comes after version 1.0.
What’s the relationship between Personal Agent Protocol and MCP?
Model Context Protocol (MCP) is one way personal agents can get work done with Personal Agent Protocol. Businesses can advertise remote MCP servers in their poppy.json file. Poppy builds on top of MCP by standardizing how agents discover businesses, sign users in with OAuth, and maintain a single session to call MCP servers, browse the web, and chat with business agents.
How can I support Personal Agent Protocol?
The first step is creating a poppy.json file and hosting it at /.well-known/poppy.json, making sure to advertise your OAuth configuration. The Personal Agent Protocol implementation kit is a great resource, and it even includes a Poppy validator that can test your implementation by acting as a personal agent.



