Brainstorm visually with Codex. Think an idea through out loud and watch it take shape on a real canvas: the mindmap, the plan, the architecture, the screens, all appearing on the MockFlow board open in your browser as you talk.
Everything Codex works out comes back as scrolling text. Anything structural, how a system fits together, where a flow branches, what the screens should be, has to be held in your head or rebuilt by hand somewhere else afterwards. This gives Codex somewhere to draw.
It works both ways. Codex reads the board too, so you can rearrange things, delete a branch, add a note, and then ask it to carry on from what is now there.
This plugin is the Codex front end for mockflow-bridge, the small local program that does the actual drawing. The bridge is required.
Codex -> this plugin -> mockflow-bridge -> your board tab
the brain the wiring the local daemon where it draws
The bridge is a program on your own computer, and it draws by talking to the MockFlow tab open in your browser. Both have to be on the same machine. So:
| Draws on your board | |
|---|---|
| Codex, selected in the desktop app | yes |
| Codex CLI | yes |
| ChatGPT in the desktop app | no, the conversation runs in the cloud |
| ChatGPT web app | no, same reason |
| Cloud tasks | no, the container is not the machine your board is open on |
ChatGPT and Codex share one desktop app, with a selector in the same interface. The selector is what matters, not the app: a ChatGPT conversation runs on OpenAI's servers even in the desktop app, so the bridge on your machine is out of its reach. A Codex session runs locally, which is why only it can draw.
The plugin still installs in ChatGPT and still carries its skills, but the board tools are not there and nothing can be drawn. It will say so and point at Codex — or at Ask Mida on app.mockflow.com for a quick board without leaving the browser — rather than quietly handing back a mermaid block.
Build boards from things only your machine can see. Point Codex at a private repo, an internal spec, a folder of meeting notes. It reads them locally and draws what it found. The files are never uploaded. Only the finished drawing is saved to your board.
It draws live, on the board you are looking at. Components appear on the open tab while you watch, saved through your own MockFlow session.
It costs no MockFlow AI credits. The thinking happens in Codex. MockFlow is the canvas, not the model.
You do not have to know what to ask for. Describe what you are trying to work out and it chooses the form: a branching flow, a board of screens, a timeline, a chart, whatever fits.
It edits instead of duplicating. Ask for a change to something already on the board and it modifies that component in place.
What you end up with is an ordinary MockFlow board. Share the link and a teammate opens it in a browser, drags things around and leaves comments. No plugin, no terminal, no idea any of it was generated.
Getting your bearings in an unfamiliar codebase.
make a mindmap of how the payments module is structured
Architecture that matches what is deployed.
draw the cloud architecture from our terraform in infra/
A schema you inherited.
read the migrations in db/ and lay out the entity relationships
Design review before anyone writes the screen.
wireframe the notification settings screen from docs/notifications.md
Incident review.
build a timeline of what happened from these log excerpts
The meeting you just came out of.
here is the transcript from this morning's planning call, turn it into a board of decisions, owners and open questions
Anything with geography in it.
map our accounts from customers.csv, then draw the three sales territories we are considering
From the Codex plugin directory: chatgpt.com/plugins/plugins_6a6cb5017a888191894b9161a0bbe0dd
The directory copy ships the skills and the /mockflow-draw command. Uploaded
plugins are currently limited to skills only, so it carries neither the bridge
connection nor any MCP tools.
That makes it guidance only, whichever surface you open it in. To draw, use Codex — selected in the desktop app, or the CLI — with the plugin installed from this repo instead:
codex plugin marketplace add mockflow/mockflow-codex-plugin
codex plugin add mockflow@mockflowThe first command registers this repo as a marketplace, the second installs the plugin from it. Check it worked with:
codex plugin list # mockflow@mockflow -> installed, enabled
codex mcp list # mockflow -> enabledThe bridge is required and is not bundled:
npm i -g @mockflow/mockflow-bridge # once
mockflow-bridge # start it, leave it runningOpen your board at app.mockflow.com, click Connect Local Agent at the top of Ask Mida, and enter the pairing code the bridge shows. One time per computer.
Requires Node.js 18 or newer and a MockFlow account.
For development, or to try a change before it is pushed, point the marketplace at a directory instead of the repo name:
git clone [email protected]:mockflow/mockflow-codex-plugin.git
cd mockflow-codex-plugin
codex plugin marketplace add .
codex plugin add mockflow@mockflowThe plugin is then read straight from that folder, so what you test is what is on disk. After editing a skill or command, reinstall to refresh Codex's cached copy:
codex plugin remove mockflow@mockflow
codex plugin add mockflow@mockflowcodex plugin marketplace upgrade only works for a marketplace added by repo
name, not a local path. To undo everything:
codex plugin remove mockflow@mockflow
codex plugin marketplace remove mockflowBoth forms register the marketplace under the same name, mockflow, so remove
one before adding the other.
Just ask, in your own words. There is no command to remember, and no need to name a component type:
draw a mindmap of how this repo is structured
Follow ups change what is already there rather than drawing a second copy:
add a Blocked column to that kanban
One command exists, for when you want certainty that the result lands on the board rather than coming back as text in the terminal:
/mockflow-draw a flowchart of the payment flow
.agents/plugins/marketplace.json the marketplace, lists the plugin below
plugins/mockflow/ the plugin itself
Inside the plugin:
| Path | Purpose |
|---|---|
.codex-plugin/plugin.json |
manifest, declares where the other pieces live |
.mcp.json |
bundled MCP server, runs mockflow-bridge stdio |
bin/mockflow-preflight |
session start check, silent when everything works |
hooks/hooks.json |
wires the preflight to SessionStart |
commands/mockflow-draw.md |
the /mockflow-draw command, only runs when typed |
skills/board-conventions |
how to use the board well, loads itself |
skills/connect |
pairing, status and troubleshooting |
assets/ |
icons used by the plugin directory listing |
Component definitions are deliberately not in here. They are fetched live from MockFlow by the bridge, so new components work without updating this plugin.
MockFlow already runs a hosted MCP server at
https://app.mockflow.com/ideaboard/mcp (used today by other AI clients). The
plugin directory's formal submission flow — the With MCP option, which
takes that URL and goes through app review — would let ChatGPT users draw
cloud boards by link from this same listing. It is deliberately not wired up:
it adds review, OAuth and rate-limit surface to maintain, and the live-board
experience through the bridge is the point of this plugin. If demand appears,
that is the path.
The same plugin exists for Claude Code at mockflow-claude-plugin. Both drive the same bridge and carry the same operating knowledge. They are separate repos because the two plugin formats differ: manifest location, MCP config shape, marketplace path, and skill frontmatter.
If you change the guidance in one, change it in the other.
MIT. Use of the MockFlow service itself is governed by the MockFlow terms of service.