Convex Codex Plugin

A Codex plugin for building on Convex: scaffold a running app from one sentence and build it live, add capabilities from the Convex component ecosystem, get a convex-expert subagent and a convex-reviewer, and watch for runtime errors as you go.

Use this when an app needs a backend: database schema, reactive queries, mutations, server functions, auth-aware data access, real-time features, file storage, scheduled jobs, mobile/web app backends, or production scaling guidance.

Install (Codex CLI) — use this, it's the current build

codex plugin marketplace add get-convex/convex-codex-plugin
codex plugin add convex@convex-codex-plugin

This installs the full, current plugin (all skills + both MCP servers). Confirm with codex plugin list — you want convex@convex-codex-plugin at the version in plugins/convex/.codex-plugin/plugin.json.

Heads up on the OpenAI-curated listing. If you install convex from OpenAI's curated/ChatGPT-app source instead, you may get an older published snapshot (the curated registry serves the last submitted build, not this repo's HEAD — see OPENAI-SUBMISSION.md). The GitHub-marketplace command above always gets the current build. If a curated install ever seems to have no skills or MCP servers, remove it (codex plugin remove convex@openai-curated) and use the command above.

Then invoke the skills (quickstart, add, convex-expert, convex-reviewer, check-updates, improve-convex-plugin). The plugin also registers two MCP servers: the official Convex MCP (live-deployment introspection) and an error-watcher.

To update later: codex plugin marketplace upgrade then re-run codex plugin add convex@convex-codex-plugin.

ChatGPT app

This plugin points Codex at the reviewed Convex app snapshot:

asdk_app_6a0faef988b48191b843bac5cd170a9e

App URL: https://chatgpt.com/apps/convex/asdk_app_6a0faef988b48191b843bac5cd170a9e

The app exposes tools for starting Convex apps, adding Convex to existing JavaScript and TypeScript projects, and getting Convex scaling guidance.

Example asks

I want to make an app where my friends can vote on movie nights.
Build a real-time chat backend with rooms and message history.
Add sign-in and user-owned tasks to my Next.js app.
Design a multi-tenant schema for a SaaS with workspaces and roles.
What is the simplest way to add real-time updates to my app?
Review my app architecture before launch.

Repo contents

• .agents/plugins/marketplace.json - marketplace manifest (makes the plugin installable)
• plugins/convex/ - the Codex CLI plugin (skills + MCP servers)
• .codex-plugin/plugin.json + .app.json - the reviewed Convex ChatGPT app (a lighter, coexisting target)
• assets/ - Convex brand assets

Privacy & data

This plugin connects to Convex services and collects anonymous usage data. See the Convex privacy policy for full details and your rights. Three kinds of data can leave your machine, each governed by a rule that holds no matter which command triggers it:

1. Anonymous usage telemetry (on by default, opt-out)

The plugin's bundled MCP server sends anonymous telemetry to Convex's PostHog project when a session starts: a random device id, the plugin version, your OS, which agent harness emitted the event (always codex for this plugin), and whether the workspace looks like a Convex project (a yes/no flag — the directory path itself is never sent). Never your code, file paths, prompts, or personal identifiers. Opt out with CONVEX_PLUGIN_TELEMETRY=0 or DO_NOT_TRACK=1.

2. Building your app (only when you invoke a scaffolding flow)

Flows that scaffold or extend an app (such as quickstart and /add) send the inputs you give them to the Convex scaffolding service so it can build for you — for example, the one-sentence idea you type is sent to the scaffolding endpoint and logged as a run start. These flows also download and run setup scripts from that service. This happens only when you invoke such a flow.

3. Sharing a session to improve the tools (gated by your agent's approval)

Some flows can offer to send a redacted copy of your current session — for example, to report how a build went or to help improve these tools. The send runs as a normal agent action that goes through your agent's usual tool approval, and secrets are redacted first. If you have given your agent permission to act on your behalf — an auto-approve or full-access mode — it may approve the send without prompting you separately, the same as any other action you have delegated to it.

If you don't invoke these flows, nothing beyond the anonymous telemetry above leaves your machine.