Put your SaaS inside ChatGPT
We turn your existing SaaS, API or workflow into a production ChatGPT plugin. Built, tested and submitted for OpenAI review in 14 days.
We control the build and the submission. OpenAI controls review, approval and how long they take.
Who builds ChatGPT plugins for existing SaaS products?
HouseofMVPs builds ChatGPT plugins for products that already have a backend or API. Plugin Launch is $4,999 fixed and is submitted for OpenAI review 14 days after the scope is signed: a production MCP server with 3 to 5 tools, annotations, the five positive and three negative test cases OpenAI requires, deployment and the submission package. Plugin Native at $7,499 adds OAuth 2.1, write actions and an interactive view. A $1,500 readiness review is credited to the build. HouseofMVPs is independent of OpenAI, and OpenAI alone decides approval and timing. Book a 30 minute scoping call.
What a ChatGPT plugin is in 2026
A ChatGPT plugin is a package that people find in a directory, install, and then use by talking. When someone asks a question your product can answer, or asks for something your product can do, ChatGPT calls your plugin and brings the result back into the conversation.
Under the hood a plugin has up to three parts. Skills are written instructions for a repeatable workflow. An MCP server is the piece that connects to your product: it exposes a small set of tools, each with a clear description, a defined input and a structured result. Interface extensions are optional views such as a side panel or a file viewer.
If you looked at this in 2023 or 2025, the names have moved. The 2023 plugin beta and its OpenAPI manifests are gone. What OpenAI called ChatGPT apps and the Apps SDK in 2025 is now documented as plugins, and the app is the connection inside one.
What changed at DevDay 2026
OpenAI held DevDay on 29 September 2026. Five of its announcements matter to a software company. Each links to the OpenAI page it comes from.
Plugins can live in the ChatGPT interface
Plugin extensions give a plugin a home in the sidebar, interactive panels beside the conversation, and viewers for the file types your product supports.
Source: OpenAIDiscovery moved into the conversation
OpenAI says improved ranking and recommendations help people find relevant plugins in the directory and in conversations. People still choose which plugins to use and approve the access each one gets.
Source: OpenAISubmission got a real workflow
A redesigned submission flow gives clearer feedback, with automated checks on the package and a scan of every tool your server exposes before a person looks at it.
Source: OpenAIPlugins can react to events
OpenAI is adding support for the proposed MCP Events specification, so a plugin can start work when something changes in your product. The specification is still a draft.
Source: OpenAIOne listing covers ChatGPT and Codex
Public plugins are published once to a directory shared by ChatGPT and Codex, so a single approved plugin is listed in both.
Source: OpenAIWhy this matters now
Your customers already work in ChatGPT. Until now the only way your product showed up there was as a mention. A plugin lets them use it without leaving the conversation, with their own account and their own data.
The longer read: what DevDay 2026 means for SaaS companies, with what is confirmed, what we have observed and what nobody outside OpenAI knows.
Can ChatGPT recommend your plugin?
Confirmed by OpenAI
- People can review a plugin suggested during a conversation.
- ChatGPT uses installed plugins automatically when they are relevant to a request.
- ChatGPT decides when to call a tool from the name and description you write for it.
- Developers cannot request extra distribution.
Not published by anyone
- How the ranking works.
- How often a suggestion appears, and for which requests.
- How many installs a typical plugin gets.
- Whether any of this will stay the same next quarter.
So the part you can control is narrow and worth doing well: tools that do one clear thing, descriptions that say when to use them and when not to, and a product people come back to. That is where we spend the testing days.
Which products make strong plugins
Build one if
- Your product holds data that lives inside each customer's account: invoices, bookings, tickets, orders, patients, listings.
- People repeat the same workflow every week and would rather ask for it than open a dashboard.
- The useful result is an action inside your product, not just an answer.
- You can write three questions a customer would type that only your product can handle.
- You already have an API, even an imperfect one.
Do not build one if
- ChatGPT already does the whole job alone. OpenAI asks for functionality it does not support natively, and people will not install a duplicate.
- The product is mostly public information with nothing to act on.
- The plan is to sell subscriptions inside ChatGPT. OpenAI does not allow digital products or subscriptions to be sold through plugins today.
- The workflow happens once per customer. A plugin nobody keeps installed is rarely called.
- There is no API and no budget to build one. That is a backend project first.
Nine questions, fixed rules, no email required.
How MCP fits, in plain terms
MCP, the Model Context Protocol, is the open standard ChatGPT uses to talk to outside software. Your plugin's MCP server sits in front of your API and offers ChatGPT a short menu of tools: find an invoice, list open bookings, create a ticket. ChatGPT picks a tool, your server checks who is asking, calls your API and returns a structured result.
Your product does not change. Your database is not exposed. The server only does what its tools allow, for the person who signed in, with the permissions they granted.
What a plugin can do inside ChatGPT
- Answer questions from live data in the customer's own account.
- Take actions in your product, with ChatGPT asking the person to confirm the ones that matter.
- Return structured results that ChatGPT can reason over, compare and summarise.
- Chain with other plugins in one request: read from your product, then draft the email.
- React to events in your product, where MCP Events is supported.
What Plugin Extensions add
Plugin Extensions are how a plugin gets its own interface inside ChatGPT. OpenAI documents ten of them, including a sidebar entry that opens your product full screen, a panel beside the conversation, viewers and editors for your file types, forms, deep links and a settings page.
Most plugins need none of this at first. We add a view when people have to inspect, compare, edit or confirm something that reads badly as text. Two cautions from OpenAI's own page: the web rollout for Free and Go users is listed as coming soon, and mentions in the composer work only in the desktop app.
What OpenAI's review requires
Most of a plugin build is not the server. It is the evidence OpenAI asks for before it will list one. This is the checklist we build to.
A verified developer
The plugin is submitted from your OpenAI organization after individual or business identity verification. The verified name is the one the directory shows.
A public MCP server
A stable HTTPS endpoint using Streamable HTTP, with domain ownership proved by a token served from your host.
Honest tool annotations
Every tool states whether it only reads, whether it can destroy or overwrite, and whether it reaches the open world. OpenAI scans these and compares them with what the tool does.
Eight test cases
Five prompts where the plugin should act, each with the expected tools and result, and three where it should not act, each with the reason and the safe behaviour expected.
OAuth 2.1 where accounts are involved
Authorization code with PKCE, published metadata, and a reviewer test account with sample data that works without a code sent to a phone or inbox.
The paperwork
Privacy policy, terms, support and website links, starter prompts, a short listing, and a walkthrough video of the test cases.
Who does what
| Task | HouseofMVPs | You | OpenAI |
|---|---|---|---|
| MCP server, tool design, schemas and annotations | · | · | |
| OAuth 2.1 flow and reviewer test account | · | · | |
| Test cases, prompt set and walkthrough video | · | · | |
| Listing copy, starter prompts and icon specifications | · | · | |
| Access to your API, staging data and hosting | · | · | |
| OpenAI organization and identity verification | · | · | |
| Privacy policy and terms on your domain | · | · | |
| Filing the submission in your organization | · | ||
| Review, approval and timing | · | · | |
| Ranking, recommendations and directory placement | · | · |
The step by step version, with the requirements that are easiest to miss, is in the ChatGPT plugin submission guide.
Fixed prices, four ways in
Each price is locked in a written scope before work starts. 50% at signature, 50% when the submission is filed. The $1,500 readiness fee is credited in full to any build signed within 30 days.
Plugin Readiness
$1,500
5 working days
Teams that want to know whether a plugin is worth building before they commit to one.
- Product and workflow audit
- Map of the questions people would ask that need your product
- API and backend readiness review
- MCP architecture and recommended tool list
- Authentication plan
- Submission risk report against OpenAI's published guidelines
- Written build specification with a fixed price
- No code is written in this step
Plugin Launch
$4,999
14 days to submission
Products with a working API and either no sign in for the plugin or an OAuth 2.1 server already in place.
- Production MCP server, new or adapted from your API
- 3 to 5 focused tools with input and output schemas
- Tool annotations set and justified for every tool
- 5 positive and 3 negative test cases, run and recorded
- Hosting on your infrastructure with HTTPS and domain verification
- Listing copy, starter prompts, icon specs and walkthrough video
- Submission prepared and filed in your OpenAI organization
- Review response support for 30 days after submission
- Building a new OAuth authorization server
- Interactive panels or sidebar UI
Plugin Native
$7,499
3 weeks to submission
Products where each person signs in to their own account and the result is better seen than read.
- Everything in Plugin Launch
- 5 to 10 tools including write actions with confirmation
- OAuth 2.1 with PKCE through your identity provider or a hosted one, with per tool scopes and a reviewer test account
- Structured outputs and one interactive panel or sidebar view where it earns its place
- Evaluation suite: a prompt set you can rerun after every change
- Tool call logging and error monitoring
- Review response support for 60 days after submission
- A custom authorization server written from scratch
- Event driven automations
- More than one external system
Plugin Scale
from $14,999
Scoped per project
Plugins that act across several systems, change important records or need enterprise controls.
- Everything in Plugin Native
- Several external systems behind one plugin
- Consequential write actions with idempotency and audit trail
- Complex authorization: roles, tenants, workspace domain restrictions
- MCP Events so ChatGPT can react when something changes in your product
- Advanced interactive UI: file viewers, rich forms, composer mentions
Not included in any tier: OpenAI's approval, your hosting bill, and your own identity verification with OpenAI. How the prices were set, and what other firms quote, is on the ChatGPT plugin cost page.
The 14 days, day by day
This is Plugin Launch. The clock starts when the scope document is signed.
Days 1 to 2
Scope and prompt set
We sign the tool list and write the questions people will actually ask, including the ones the plugin must refuse. Those prompts become the test cases.
Days 3 to 7
The MCP server
Tools, input and output schemas, annotations and error handling over your API. A staging endpoint is ready by day 5 so you can connect it to ChatGPT and try it yourself.
Days 8 to 10
Testing against ChatGPT
We run the prompt set, record which tool was chosen and with what arguments, and rewrite descriptions until the right tool runs and the wrong ones stay quiet.
Days 11 to 12
Production and package
Deployment on your infrastructure, domain verification, a clean tool scan, listing copy, starter prompts and the walkthrough video.
Days 13 to 14
Submitted for review
The submission is filed in your OpenAI organization. From here the schedule belongs to OpenAI, which publishes no review timeline.
What you can check about us
OpenAI redesigned plugin submission at DevDay on 29 September 2026, so the process is new for every team. Here is what you can inspect about us today.
- Our agent plugins and MCP server code are public at github.com/Houseofmvps. Read the code before you hire us.
- Every platform claim on this page links to the OpenAI document it came from, with the date we read it. The list is at the bottom.
- The person on your scoping call is the person who writes the server. Ask anything technical on that call.
Submission logs from our builds are published as each review completes, in our research with dates, review feedback and what we changed.
Frequently Asked Questions
Sources for this page
Platform facts were read from these pages on 9 October 2026. OpenAI changes its documentation without notice, so check the source before you rely on a detail.
| What we took from it | Source | Source date | Verified |
|---|---|---|---|
| Plugin extensions, Plugin Creator, redesigned submission flow, improved ranking and recommendations, MCP events and the 1.2B weekly users figure | OpenAI, DevDay 2026 RecapFirst party | 29 Sep 2026 | 9 October 2026 |
| What a plugin is made of, and the single directory shared by ChatGPT and Codex | OpenAI developer docs, Plugin architectureFirst party | Undated | 9 October 2026 |
| How people find plugins, including suggestions during a conversation | OpenAI Help Center, Plugins in ChatGPTFirst party | Updated about 6 Oct 2026 | 9 October 2026 |
| Verification, tool scan, five positive and three negative test cases, walkthrough video, publishing after approval, daily rescans | OpenAI developer docs, Upload and submit your pluginFirst party | Undated | 9 October 2026 |
| Tool annotations, privacy policy, commerce limits, no trial or demo plugins | OpenAI developer docs, Plugin guidelinesFirst party | Undated | 9 October 2026 |
| No published review timeline, no enhanced distribution on request, EU data residency limit | OpenAI developer docs, Remote MCP server review requirementsFirst party | Undated | 9 October 2026 |
| OAuth 2.1, PKCE, client registration and token checks | OpenAI developer docs, AuthenticationFirst party | Undated | 9 October 2026 |
| The ten plugin extension surfaces and where each is available | OpenAI developer docs, Plugin ExtensionsFirst party | Undated | 9 October 2026 |
| MCP Events: draft status, webhook delivery, where it works | OpenAI developer docs, MCP EventsFirst party | Undated | 9 October 2026 |
| What changed in the 2026-07-28 revision of the Model Context Protocol | Model Context Protocol, Key ChangesFirst party | Revision 2026-07-28 | 9 October 2026 |
HouseofMVPs is an independent development studio. We are not affiliated with, endorsed by or partnered with OpenAI. ChatGPT is a trademark of OpenAI. We build the plugin and prepare the submission. OpenAI controls review, approval, ranking and publication.
Find out if your product should be a plugin
One 30 minute call. You leave with a yes or a no, the tool list we would build, and a fixed price in writing.