Can you turn the clients and follow-ups we just worked out into an app I can keep?
openmcp.app open_app Done
The chat ends. Your app doesn’t.
Done. I turned this conversation into a working client pipeline — already filled, shared with your AI, and still here next time.
You and your AI keep updating the same data. Reopen it in a new conversation, another AI, or at its own link.
Built from your brief
Not a template you have to adapt.
Your words became the App contract, the useful views, and the review-before-send boundary for this workflow.
- Describe: Clients · amounts · next actions
- Build: Fields · views · safe actions
- Keep: App ID · data · version history
Client Pipeline
Four clients from this conversation, organized into a live pipeline with overdue work and next actions already visible.
4 clients from this chat · One state for you + AI · Same app next time
Open Client PipelineBuild one free See the shared state
What did we finally decide the invoice payment terms should say?
It’s here: net 30 starts from the invoice date unless the contract says otherwise. I opened the Knowledge Cards app where we kept that answer, alongside the wording you later marked as the version to reuse.
Knowledge Cards
Eight useful answers from recent conversations. The net-30 rule and your current invoice wording sit beside a user photo, a credited web reference, and a clearly labeled AI reference.
8 saved answers · 3 labeled image sources · Your wording is current
Open Knowledge CardsYou retitled a card to “Invoice wording — use this one” · just now
I see the edit. “Invoice wording — use this one” is now the version I’ll pull when another app needs it.
Three days later · new conversation
Can you pull up the invoice wording we decided on?
Here it is — in the original Knowledge Cards app, together with the two answers you added later. Nothing was reconstructed for this conversation.
Knowledge Cards
The original answer library reopened with the net-30 card, your renamed invoice wording, and later additions intact.
Same answer library · Later additions included · Original edit retained
Open Knowledge Cards- The useful answers were kept: Eight answers became one searchable App, with user, web, and AI images retaining their real provenance.
- One wording became the current version: You renamed the card so the AI could tell which invoice wording to reuse.
- The same library reopened: The net-30 answer, renamed wording, and two later additions returned together.
Who’s overdue — and can you get the chasers ready?
Two clients are overdue. I used “Invoice wording — use this one” from Knowledge Cards, prepared two Gmail drafts — unsent for review — and logged both follow-ups on the correct client rows.
Client Pipeline
Four open clients, two overdue, with the saved wording reused in two reviewable drafts and both attempts logged against the right accounts.
$15,400 open · 2 overdue · 2 drafts ready
Open Client Pipeline- Invoice wording — use this one: The version you confirmed became the source for both follow-up drafts
- Nakamura Studio: Correct invoice and due date included · unsent for review
- Marrow & Field: Prepared with the correct invoice and due date · unsent
- Two follow-ups recorded: The action and its context now live on the client rows
Knowledge Cards read · 2 Gmail drafts ready · unsent · 2 follow-ups logged
Knowledge Cards supplied the wording, Gmail supplied the action, and Client Pipeline kept the result, context, and history.
Do I always have to start from scratch?
Start ready-made. Make it yours.
The App Store is an app too: search working apps, open a live preview, add one, then reshape it by asking.
This is the real hosted catalog — not a gallery of screenshots.
Explore the full App Store
I just changed Kite & Co’s next step in ChatGPT. If I open Claude tomorrow — or open the app with no chat at all — am I looking at the same work?
Change the window. Keep the work.
You are not moving a document between AIs. ChatGPT, Claude, and the app’s own URL all reopen client-pipeline: one App ID, four client rows, and one version history. A write in one surface becomes the starting state in the next.
- ChatGPT: Writes to the app while you talk — Set Kite & Co to Friday follow-up; New next step saved
- Claude: Starts a fresh conversation — What comes next for Kite & Co?; Reads Friday follow-up from the app
- Web app: Opens without an AI conversation — Review the pipeline and prepared drafts; Same four rows and history
Client Pipeline
The same live pipeline and rows underneath every surface.
Same 4 client rows · Edits carry forward · One version history
Open Client PipelineIf the App already holds the same four clients and our decisions, do I still need to paste all of that into every new AI conversation?
Keep the structure. Send the change.
The first use establishes the App contract and initial state. Later use reopens the same App ID, reads the current rows it needs, and can update the changed field instead of rebuilding the whole table. That can reduce repeated context and leave more of the model’s working window for the next decision.
First use · establish
Build the reusable working surface
The useful structure and initial state are created once, then persist beyond the chat.
- App contract
- Fields · views · allowed actions
- Initial state
- 4 client rows · current wording
- Rendered App
- One shared, inspectable surface
Next use · reuse
Open current state, then patch the delta
The same schema and unchanged rows do not need to be narrated from scratch.
- Reuse
- Same App ID + contract
- Read
- Current row when needed
- Update
- One changed field + new version
Local update · v18 → v19 · one row changed
Kite & Co
next_stepFollow up Friday → Proposal reviewed · follow up Monday
- Less repetition: Stop pasting the same background into each new chat.
- Less rebuilding: Reuse the App instead of regenerating a table or dashboard.
- More useful context: Spend more of the model’s context on the current decision and result.
This is a mechanism, not a fixed savings claim. Actual token use depends on the AI host, model, tool-call format, and workflow; an App does not create unlimited memory or cache every conversation.
I want to use it today, but I also want the option to run it myself later. What do I give up — and what responsibility do I take on?
Start hosted. Keep the right to run it yourself.
These are not two products with different features. App manifests, collections, and version history use the same AGPL-3.0 engine. The fork is operational: who runs it, where data lives, who updates it, and which MCP endpoint your AI opens.
Shared foundation
open-mcp-apps · AGPL-3.0
The same engine, App manifests, collections, and version history model.
Ready in minutes
Hosted
Free during the public beta. Create an account, connect your AI, and use a separate hosted store for your apps without installing a server.
Create account → Add the managed MCP URL → Open your apps
- Operator
- openmcp.app
- Storage
- Your account store
- MCP endpoint
- Managed URL
- Updates
- Managed by us
No server setup · One store per account · Free in public beta
Start freeThe open engine
Self-host
Run the AGPL-3.0 engine on your own machine or server. You control deployment, storage, updates, and the MCP endpoint.
Run the open engine → Choose storage and a URL → Connect your AI
- Operator
- You or your team
- Storage
- Your infrastructure
- MCP endpoint
- Your URL
- Updates
- You decide and run
AGPL-3.0 · You operate updates · You choose the infrastructure
View on GitHubPrivate deployment
Company
For several people, a private environment, or deployment requirements that need planning beyond a personal install.
Deployment review → Private environment → Planned rollout
Several people · Private requirements · Human support
Talk to usMoving hosted data to self-hosting currently requires an export request; self-serve export and migration are not built yet. Hosted pricing after the public beta is also undecided. We will state both before anything changes.
Before I put real work in it, tell me exactly where the boundaries are.
Know the boundary before you rely on it.
What is available today, what depends on the AI host, and what is still plainly not built.
- Data · Hosted or self-hosted · Where does my data live?
- Hosted accounts use one store per account on our servers. Self-hosted installs keep it on your machine. When an AI opens an app, that data also passes through that AI provider under its privacy terms.
- Continuity · Open engine · What happens if openmcp.app shuts down?
- The engine is open source under AGPL-3.0, so you can run it yourself. Data export is available by request today; a self-serve export button is not built yet.
- Host support · Works with limits · Does it work in Claude and ChatGPT?
- Claude and ChatGPT web can render the apps. ChatGPT desktop can render them too, but a button inside an app cannot create new rows there yet because of a host limitation.
- Current gaps · Not built yet · What can’t it do yet?
- Apps are personal for now: there is no teammate sharing and no public app you can hand someone a link to. Both are on the roadmap, and sharing will be a hosted-account feature — a self-hosted install has no one to share with. Export is by request, and ChatGPT desktop still has the write limitation above. Every app version is kept so you can roll back a bad change.
