Work on Onvoi apps from your coding tools
Apps your organization builds on Onvoi can be changed from Claude Code, Cursor or any Git client. Ask Onvoi's coding agent through the MCP connection, or clone the code and push a branch yourself.
1. Through the Onvoi connection
If you already connected Onvoi (see Connect your AI tools), the app tools are included. Connected before these existed? Disconnect in Manage connections and connect again to approve app access. In Claude Code:
$ claude mcp add --transport http onvoi https://api.onvoi.ai/api/mcp
Then just ask, for example:
app.listapp.readapp.change_request.createapp.change_request.readapp.change_request.replyapp.deploy.request2. With Git
Every app has a Git repository. Find its clone URL on the app’s Source & advanced page. Git asks for a username and password: type any username, and paste an Onvoi Git token as the password — not your Onvoi password.
$ git clone https://git.onvoi.ai/<org>/<app>.git $ git checkout -b my-change # edit, commit, then $ git push origin my-change # builds a preview
Already built it somewhere else?
Create the app with Develop with Claude Code or Codex and run the one setup command it shows — or use the plain Git commands below. Your agent can then push, read results and fix problems itself: see Build with Claude Code or Codex and Push an existing project.
Quick setup — if you’ve done this kind of thing before
Built it in Claude Code, Codex or anything else? Push it here from your terminal. Read the guide.
git init git add -A git commit -m "first commit" git branch -M main git remote add onvoi https://git.onvoi.ai/<your-workspace>/<your-app>.git git push -u onvoi main
git remote add onvoi https://git.onvoi.ai/<your-workspace>/<your-app>.git git push -u onvoi main
Tip: let Git remember it — git config --global credential.helper osxkeychain on a Mac, or store elsewhere.
No onvoi.yaml? Onvoi sets one up from your code and shows it to you before anything goes live. An admin’s push updates main right away; going live always waits for an approval.
The app’s settings file
Each app has an onvoi.yaml at the top of its repository: what it runs, its data, who can do what, and the secrets it needs (by name only — values are entered in Onvoi). Most apps run on Onvoi’s shared setup at no fixed cost; a dedicated server is only used when someone turns it on under Advanced → Runtime.
apiVersion: onvoi.ai/v1alpha1
kind: Application
metadata:
name: inventory
license: MIT
runtime:
stack: static
functions:
- name: restock
handler: functions/restock.handler
route: { method: POST, path: /api/restock }
auth:
mode: onvoi
services:
database: { type: mongodb, provision: true }
secrets:
- name: SUPPLIER_API_KEYBuilding with Claude Code, Codex or Cursor?
Every Onvoi app repository includes an onvoi-code.md: the conventions Onvoi’s own coding agent follows — the settings file, the platform rules, and the default look. You don’t have to point your tool at it: the repo also carries an AGENTS.md (read automatically by Codex, Cursor and most coding agents) and a CLAUDE.md (read automatically by Claude Code) that send it there, so what it builds deploys on Onvoi without fixes. Connected to the Onvoi MCP? app.read returns the same conventions. Starting a project outside Onvoi? Download onvoi-code.md, or:
$ curl -o onvoi-code.md https://api.onvoi.ai/api/app-platform/onvoi-code.md
Apps from the marketplace
Installing an app is one decision: its first version goes live in your workspace without a separate approval. From then on there are two ways to own it:
- Use the original — your copy follows the original. Changes you ask for are suggested to its makers instead of published to your copy; if they accept, you get them with the next update. Choose Make it my own version any time to change it yourself.
- Make it yours — your own version from the start. Give it your own name (rename it from its page), publish it, and others can suggest changes to it.
Ratings and the agent score
Teams that use an app can rate it (stars and a few words). Separately, every app gets an agent score from 0 to 10: how easily people’s AI agents use its tools, based on how their calls go. It appears on the listing once enough teams’ agents have used the app. The people who look after the app see the score privately, with suggestions for what would make it easier for agents — each one a click away from a change request. Clear capability descriptions and input schemas are what raise it.