# Personal Agent Protocol vs MCP: what each one does, and why you will run both · Stacktree

Source: https://stacktr.ee/personal-agent-protocol-vs-mcp

[Skip to content](#main) PFrom Product Hunt? Your first pages are free, no card. Code `PRODUCTHUNT` takes 50% off your first month of Studio.[Start free &rarr;](https://app.stacktr.ee/?join=1&from=producthunt)

&times;   A Stacktree customer sent you. Your first month is 50% off on any plan, taken off at checkout. [Start free &rarr;](https://app.stacktr.ee/?join=1&from=referral)

 &times;    [Stacktree](/)[Developers](/developers)[Agents](/agents)[Docs](/docs)[Use cases](/use-cases)[Pricing](/pricing)[Blog](/blog)[Dashboard](https://app.stacktr.ee)[Sign in →](https://app.stacktr.ee)      By [ Steve Smith ](/about) · Founder, Stacktree  ·  Last updated November 15, 2026          Personal Agent Protocol · MCP

#  Personal Agent Protocol vs MCP.

   They are not rivals. MCP is how an agent calls your tools. The Personal Agent Protocol, Poppy for short, is how an agent you have never heard of finds your company, proves who it is and gets its person's approval before that call. Stacktree serves both on one endpoint, so this is the comparison with the wiring showing.

  [ Watch both on one request ](https://poppy-simulator.stacktr.ee)    [ The protocol, explained ](/personal-agent-protocol)  The Poppy Simulator calls our MCP server over a Poppy Session.

##  What is the difference between the Personal Agent Protocol and MCP?

MCP (Model Context Protocol) standardizes how an AI agent lists a server's tools and calls them. The Personal Agent Protocol (Poppy) standardizes what comes before the call when the agent belongs to a member of the public: finding the company from its domain, proving which agent is asking, tying requests to one person, and getting that person's approval on the company's own page. Poppy lists MCP as one kind of API a company can offer, so most companies with an MCP server keep it and add a `poppy.json` that points at it.

## On this page

  -   01  [ Side by side ](#table)
-   02  [ What the spec says about MCP ](#spec)
-   03  [ One endpoint, both protocols ](#wire)
-   04  [ What Poppy adds that MCP leaves open ](#adds)
-   05  [ What MCP does that Poppy does not ](#mcp-only)
-   06  [ Which do you need? ](#which)
-   07  [ Three things that bite when you add Poppy to an MCP server ](#gotchas)
-   08  [ And UCP? ](#ucp)

## Side by side

 The Poppy column is from [the 1.0 spec](https://personalagentprotocol.org/docs/1.0/spec), published 14 November 2026.

    Question MCP Personal Agent Protocol     What does it standardize? Listing a server's tools and calling them How a person's agent and a company meet: discovery, identity, Sessions, sign-in   How does an agent find you? It is given your server's address, by a person or a directory It fetches `/.well-known/poppy.json` from your domain   How does the client say who it is? Through MCP's OAuth flow, when the server asks for sign-in On every request, signed out too: a signed `Poppy-Session` header tied to a metadata URL it publishes   Who is the person? Whoever signed in A stable, pseudonymous User ID per company before sign-in, and the account after   What can the agent reach? The MCP server APIs (MCP or OpenAPI), your website in a browser, and your own support agent   Who defines the tools? MCP does: schemas, results, errors Nobody. The spec "leaves each API's schema to that API"   When you cannot help? A tool error A list of tasks you chose not to offer, and a feedback endpoint for what was asked

## What the spec says about MCP

 Section 7 of the spec defines two API types a company can list: `openapi` and `mcp`. The MCP rules are short:

  - Which MCP. A remote server on the Streamable HTTP transport, MCP version 2025-06-18 or later.
 - Sessions are separate. "Poppy Sessions are independent of MCP's transport lifecycle." Whether the server uses the older initialize handshake or the per-request metadata of [MCP 2026-07-28](/blog/mcp-2026-spec-changes), the agent sends the same Poppy-Session header. The guides add that the Session's id "is not an Mcp-Session-Id".
 - Signed out works. A server that accepts the Session header can take signed-out calls with the header only, if it allows them.
 - Tokens are for one server. A signed-in agent asks for a token with resource set to the MCP server's address from poppy.json. That token works only there.
 - Other MCP clients are untouched. Clients that do not implement the protocol "sign in with MCP's own authorization flow, outside this protocol, and their requests belong to no Session".

## One endpoint, both protocols

 Here is the entry in [our live poppy.json](https://stacktr.ee/.well-known/poppy.json), with the description shortened:

```
`"apis": [
  {
    "type": "mcp",
    "url": "https://api.stacktr.ee/mcp",
    "description": "Publish a page to a private link that opens with no account, update it in place, see who opened it and what they commented."
  }
]`
```

 And here is a personal agent that has never contacted us before, listing our tools signed out:

```
`POST https://api.stacktr.ee/mcp
Poppy-Session: eyJ0eXAiOiJwb3BweS1zZXNzaW9uK2p3dCIs...
Poppy-Version: 1.0
Content-Type: application/json

{ "jsonrpc": "2.0", "id": 1, "method": "tools/list" }`
```

 That is an ordinary MCP request with two extra headers. The same address serves the Claude and ChatGPT connectors, which send neither header and sign in through MCP's own OAuth flow. Three things are shared between the two paths: the tools, the two scope names (`sites:read` and `sites:write`), and the consent page a person sees. One middleware in front of the server checks the Session header when there is one.

## What Poppy adds that MCP leaves open

  - Discovery from a domain. A person tells their agent "return this to Example" and the agent starts from example.com. Nobody has to install a connector first.
 - An identity for agents nobody registered. The agent's client_id is the address of a metadata document with its name and keys (section 4.1). You can tell a customer's agent from a scraper on the first request, and block or rate-limit one agent without touching the rest.
 - A person before sign-in. The User ID is stable for one person at one company and different at every other company (5.1). You can keep a cart or a draft for a signed-out person and still not know who they are.
 - Rules for the consent page. It is your page, it names the agent, and since 1.0 it lists every scope every time. Device Sign-In has its own rules for agents with no browser.
 - The website and your own agent. A browser can join the Session and use your site signed in without the token passing through it (section 6). A personal agent can talk to your support agent and be handed to a person (section 8).
 - A way to say no. Tasks you chose not to offer, and feedback when a person asked for something you cannot do (section 9).

## What MCP does that Poppy does not

  - The call itself. Tool schemas, results, resources, prompts, and asking the person a question in the middle of a tool. Poppy has none of this and does not try to.
 - Clients people already use. Claude, ChatGPT and most coding agents connect to MCP servers today. A person adds your server once and uses it from then on. Poppy is five weeks old, and Sierra says its first large agent, Meta's Muse, switched it on this month.
 - Private and internal use. An MCP server can sit behind a company's firewall for its own staff. Poppy is about the public: a company and the agents of people it has never met.

## Which do you need?

  - You have an MCP server. Add a poppy.json that lists it. For signed-out use that is one static file and letting two headers through. Our [field guide](/poppy-json) has the smallest valid file.
 - You have a REST API and no MCP server. You do not need MCP for Poppy. List your OpenAPI description instead. Build an MCP server when you want to be in Claude or ChatGPT, which is a separate decision.
 - You have neither. Ask who you want to serve first. People who will set up a connector once and keep it: MCP. Agents that arrive at your domain unannounced on a person's errand: Poppy, and its web channel can be the whole of your first version.
 - You build agents. Speak both. Poppy tells you where a company's MCP server is and how its person signs in. MCP is still how you call it.

## Three things that bite when you add Poppy to an MCP server

 All three are from our own move, written up in [our implementation report](https://github.com/personal-agent-protocol/personal-agent-protocol/issues/75).

  - Protected resource metadata on a shared host. Our MCP server and our REST API share api.stacktr.ee. The metadata document at the host's root named the MCP server as its resource, because that is where MCP clients were sent. A personal agent reads that same root document before calling the REST API, and would have asked for the wrong token. The root document has to name the host. The MCP server gets its own document at /.well-known/oauth-protected-resource/mcp.
 - One token does not fit both APIs. A token asked for with the MCP server as its resource must be refused everywhere else. If your REST API and MCP server accepted the same token before, they cannot for personal agents.
 - Do not break your connectors. The authorization server is shared, so every Poppy change is a risk to Claude and ChatGPT sign-in. We left DPoP out of our issuer's metadata for that reason, which 1.0 allows, and we keep a test that runs the connectors' OAuth flow with Poppy switched on.

## And UCP?

 The Universal Commerce Protocol covers catalog search, carts and checkout. Poppy has an optional UCP extension, co-written with Shopify according to Sierra, that connects its discovery and sign-in to UCP. It shows in the wild: every Shopify store we checked on 15 November lists a UCP endpoint in its `poppy.json` and no MCP server. So the pattern is the same for commerce as for tools. Poppy is the front door, and the protocol behind it depends on the job. [Our tracker](/who-supports-poppy) has the files we read.

FAQ

## Frequent questions

     Does the Personal Agent Protocol replace MCP? +  No. The Personal Agent Protocol names MCP as one of the two kinds of API a company can list, next to OpenAPI. It does not define tools or how to call them. A company with an MCP server keeps it and adds a poppy.json that points at it.   What is the difference between Poppy and MCP in one sentence? +  MCP standardizes the call: how an agent lists a server's tools and runs one. Poppy standardizes the meeting: how an agent finds a company from its domain, proves which agent it is, acts for one person, and gets that person's approval on the company's own page.   I already have an MCP server. What do I need to add for the Personal Agent Protocol? +  Publish /.well-known/poppy.json on your main domain with an apis entry of type mcp that points at the server, and let requests that carry a Poppy-Session header through. That covers signed-out use. For account access, your authorization server also has to accept agents identified by a metadata URL.   Does an MCP server have to verify the Poppy-Session header? +  It has to accept the header, not always verify it. A server that uses nothing from the Session can serve signed-out calls or rely on an access token alone. It must verify the header before it relies on the Session for identity, ownership or state, such as a cart or a conversation.   Is a Poppy Session the same as an MCP session? +  No. The spec says Poppy Sessions are independent of MCP's transport lifecycle, and the guides say the Session's id is not an Mcp-Session-Id. A Poppy Session is one person's activity at one company through one agent, across APIs, the website and conversations.   Can Claude or ChatGPT connectors still use my MCP server after I add Poppy? +  Yes. MCP clients that do not implement the Personal Agent Protocol sign in with MCP's own authorization flow, and their requests belong to no Session. On Stacktree one endpoint serves the Claude and ChatGPT connectors and personal agents, with the same two scopes and the same consent page.   Which MCP versions does the Personal Agent Protocol work with? +  Version 1.0 asks for a remote MCP server on the Streamable HTTP transport, MCP version 2025-06-18 or later. It covers both lifecycles: the older initialize handshake, and the per-request metadata of MCP 2026-07-28. Both carry the Poppy-Session header the same way.   How does the Personal Agent Protocol relate to UCP? +  UCP, the Universal Commerce Protocol, covers catalog search, carts and checkout. Poppy has an optional UCP extension that connects its discovery and sign-in to UCP. The Shopify stores we checked list a UCP endpoint in poppy.json and no MCP server.
Keep reading

## Related guides

   -  [ The Personal Agent Protocol, explained Version 1.0 step by step, what changed from the drafts, and what it took to run it. ](/personal-agent-protocol)
-  [ poppy.json, field by field Where the MCP entry goes, with a live file and the smallest valid one. ](/poppy-json)
-  [ How to implement it as a company The order to build in, what each stage costs, and how to test against an agent you did not write. ](/implement-personal-agent-protocol)
-  [ MCP spec changes in 2026 The per-request lifecycle that 2026-07-28 introduced, and what else changed. ](/blog/mcp-2026-spec-changes)

References

## Sources and further reading

   -  [ Personal Agent Protocol 1.0, section 7: APIs ↗ The API types a company can list, and the rules for MCP servers. ](https://personalagentprotocol.org/docs/1.0/spec#apis)
-  [ Personal Agent Protocol guides: MCP and UCP interoperability ↗ Lifecycles, tokens and caching when one deployment speaks more than one protocol. ](https://personalagentprotocol.org/docs/1.0/guides#both-interoperability)
-  [ Model Context Protocol specification, 2025-06-18 ↗ The oldest MCP version a Poppy company may list. ](https://modelcontextprotocol.io/specification/2025-06-18)
-  [ Model Context Protocol specification, 2026-07-28 ↗ The current MCP version, with per-request metadata. ](https://modelcontextprotocol.io/specification/2026-07-28)
-  [ RFC 9728: OAuth 2.0 Protected Resource Metadata ↗ The document both protocols use to point a client at the sign-in server. ](https://www.rfc-editor.org/rfc/rfc9728)
-  [ Stacktree implementation report, Version 0.3 ↗ Our write-up of serving MCP connectors and personal agents from one endpoint. ](https://github.com/personal-agent-protocol/personal-agent-protocol/issues/75)

##  One MCP server, reachable both ways.

 Stacktree turns what an agent makes into a private page its person can share. Add it as a connector in Claude or ChatGPT, or let a personal agent find it from our domain. Free to start, no card.

 [ Sign up free → ](https://app.stacktr.ee/?join=1&from=cta_personal_agent_protocol_vs_mcp)       Private hosting for the HTML your agents make.

[](https://betalist.com/startups/stacktree?utm_campaign=badge-stacktree&utm_medium=badge&utm_source=badge-featured)[Featured on](https://devhunt.org/tool/stacktree)[](https://buildlist.io)[Get started free](https://app.stacktr.ee/?join=1)

## Product

- [Page video](/page-video)
- [Ask this page](/ask-this-page)
- [Client spaces](/client-spaces)
- [Templates](/templates)
- [Example deliverables](/examples)
- [Custom domains](/custom-domains)
- [Client feedback](/client-feedback-loop)
- [One-time links](/one-time-view-links)
- [See how pages get read](/page-engagement)
- [Made with Stacktree](/made-with)
- [Security](/security)
- [Watch the demo](/demo)

## Agents

- [All integrations](/agents)
- [Claude Code](/claude-code)
- [OpenAI Codex](/codex)
- [Cursor](/cursor)
- [Claude.ai connector](/claude-ai-connector)
- [Claude Cowork](/what-is-claude-cowork)
- [Meta Muse](/muse)
- [MCP server](/mcp-publish-html)
- [Deploy from Claude Code](/deploy-html-from-claude-code)
- [Skills](/skills)
- [Slack app](/slack)
- [n8n node](/n8n)
- [Agent payments (x402)](/x402)

## Alternatives

- [All comparisons](/alternatives)
- [Head-to-head comparisons](/compare)
- [Tiiny Host](/tiiny-host-alternative)
- [GitHub Pages (private)](/github-pages-private-alternative)
- [Vercel](/vercel-alternative-for-agents)
- [ngrok](/ngrok-alternative-for-html)
- [Display.dev](/display-dev-alternative)
- [Static.app](/static-app-alternative)
- [OpenAI Codex Sites](/openai-codex-sites-alternative)
- [here.now](/here-now-alternative)
- [Shippage](/shippage-ai-alternative)
- [Best private hosting](/best-private-html-hosting)

## Use cases

- [All use cases](/use-cases)
- [Share with clients](/share-with-clients)
- [Send a file to a client](/share-html-file-with-client)
- [Share Claude artifacts](/share-claude-artifacts)
- [Share Jupyter notebooks](/share-jupyter-notebook-html)
- [Host Storybook privately](/host-storybook-privately)
- [Architecture diagrams](/share-architecture-diagrams)
- [AI-generated reports](/host-ai-reports)
- [Internal HTML tools](/internal-tool-hosting)
- [Private HTML hosting](/private-html-hosting)
- [Vibe-coded page hosting](/vibe-coding-hosting)
- [Leave a website builder](/website-builder-migration)

## Learn

- [Blog](/blog)
- [Glossary](/glossary)
- [FAQ](/faq)
- [Where AI output should live](/where-ai-output-lives)
- [Agent-loop hosting](/agent-loop-hosting)
- [Why agents need a publish primitive](/blog/why-agents-need-a-publish-primitive)
- [MCP servers explained](/blog/mcp-servers-explained-for-developers)
- [What changed in the 2026-07 MCP spec](/blog/mcp-2026-spec-changes)
- [Sites in Codex explained](/blog/sites-in-codex-explained)
- [Private-by-default hosting](/blog/private-by-default-html-hosting)
- [An agent paid us $1 (x402)](/blog/agent-paid-to-provision-itself)
- [When a loop hits a paywall](/blog/loop-engineering-paywall)
- [Pricing](/pricing)
- [Self-host (new)](/self-host)
- [Changelog](/changelog)
- [Docs](https://stacktr.ee/docs)
- [About](/about)
© 2026 Stacktree · stacktr.ee

[Privacy](/privacy)[Terms](/terms)[Security](/security)[Dashboard](https://app.stacktr.ee)[npm](https://www.npmjs.com/package/stacktree-mcp)[Sitemap](/sitemap.xml)[llms.txt](/llms.txt)[API spec](/openapi.json)

---
Full markdown summary of the Stacktree marketing surface: https://stacktr.ee/llms-full.txt
