Agents · Codex
Let Codex design your screens, then build them
Updated
Connect the Codex CLI to Saymock and it gets a design board: it lays out screens as HTML + Tailwind, checks them with screenshots, and your team reviews them in the browser before a line of product code is written. When the design is approved, Codex exports it and builds it in your repository.
What Codex gets
- Screens it can write in one call. A frame is HTML with Tailwind classes, the markup Codex already writes; repeated rows are written once as a template.
- Edits by element. Every element keeps a stable id, so Codex changes one button instead of rewriting the screen, which saves tokens.
-
Checks.
get_screenshot,check_layoutandcompare_to_referencelet it see and fix its own work. - A review loop with people. Comments on the board become tasks Codex picks up, changes, checks and answers.
-
Code out.
export_codereturns React, Vue or HTML files with the same classes, which Codex then fits to your components.
Set it up
-
Add the server:
codex mcp add saymock --url https://app.saymock.com/mcp -
Codex detects the OAuth sign-in and opens Saymock in the browser (if it does not, run
codex mcp login saymock). Sign in or create a free account, and approve. - Start Codex and ask it to design a screen.
For a script or a setup without a browser, create a personal access token in Saymock
(Settings → Connect your agent) and put it in ~/.codex/config.toml:
[mcp_servers.saymock]
url = "https://app.saymock.com/mcp"
http_headers = { Authorization = "Bearer smk_…" }
A token is a password: keep it out of committed files.
Things to ask
Design a sign-up flow for this app in Saymock: email, verification code, profile. Mobile, three screens.
Handle the comments on the “Settings” board in Saymock, then tell me what you changed.
Export the approved screens from Saymock and implement them in our Next.js app.
Questions
Does Saymock call OpenAI or any model?
No. Codex does the work with your own account; Saymock stores, renders and shares the board.
How much does it cost?
Agents are free and unlimited: Saymock bills human editors only, 2 of them free per workspace. See the pricing.
Is the result a picture or code?
Code. Every screen is HTML with Tailwind v4 classes and no scripts: people edit it on the board, and Codex exports it to React, Vue or HTML.
How do I check the connection?
Ask Codex to call whoami: it names your account and workspaces.
Other agents: Claude Code, Cursor, ChatGPT, Gemini CLI, all of them. What to ask for: screens from your code, design to code, all use cases.