Promptify MCP

Your workspace, your credits

MCP (Model Context Protocol) lets your assistant use Promptify’s tools. It can list and read your ads, check credits, browse templates, generate scripts and start video renders. Paid actions use your Promptify credits; connecting does not grant extra credits. Credit checks, quotes and browsing tools do not deduct credits.

Credit packs: $5 for 1,150 credits, $20 for 2,000, or $50 for 5,500. Monthly plans: Creator $149, Pro $299 and Agency $799. See pricing for the current offer. Out-of-credits tool errors include that link.

Connect with OAuth

Claude, ChatGPT and Codex clients that support remote MCP OAuth can connect to https://www.promptify.pro/api/mcp. The client discovers Promptify’s authorization server and registers itself, or uses its client metadata document. Sign in to Promptify, review the named assistant and workspace, then approve or deny access.

Approval lets the assistant read workspace data and generate scripts using your credits. Rendering still requires you to confirm a fresh credit quote. API keys remain available for header-based clients.

Get your API key

1. Sign in to Promptify.
2. Open Settings → API and create a named key.
3. Copy the pfy_live_… key immediately. It is shown once. Store it privately; revoke a lost or exposed key and create a replacement.

Use Streamable HTTP with this URL and header:

https://www.promptify.pro/api/mcp
Authorization: Bearer pfy_live_…

Replace the placeholder with your own key. Keep keys out of chat messages and shared configuration files.

Claude Desktop and Claude.ai

In Claude’s connector settings, add the remote MCP URL. A client that supports custom headers needs the Authorization header above. This is the URL-and-header configuration for remote-capable clients:

{
  "mcpServers": {
    "promptify": {
      "url": "https://www.promptify.pro/api/mcp",
      "headers": {
        "Authorization": "Bearer pfy_live_…"
      }
    }
  }
}

Claude’s OAuth-capable connectors can open Promptify’s sign-in and consent flow from the remote URL. For a Desktop setup that uses an API key, the third-party mcp-remote bridge is another option (Node.js required). Add this configuration and restart Claude:

{
  "mcpServers": {
    "promptify": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://www.promptify.pro/api/mcp",
        "--header", "Authorization: Bearer pfy_live_…"
      ]
    }
  }
}

ChatGPT, OpenAI apps and Codex

Add the remote URL where your ChatGPT account supports MCP and select OAuth when configuring authentication. Complete Promptify sign-in and consent in the browser. OAuth-capable Codex clients use the same discovery and consent flow. Client availability and setup controls depend on your account and client version.

For an OpenAI API app, configure a remote MCP tool with server_url set to the endpoint and authorization set to your Promptify key (the token only, without the Bearer prefix). Keep tool approvals enabled. See the official remote MCP guide.

For Codex, set PROMPTIFY_API_KEY in the environment that launches Codex to your key (without the Bearer prefix), then add this to ~/.codex/config.toml:

[mcp_servers.promptify]
url = "https://www.promptify.pro/api/mcp"
bearer_token_env_var = "PROMPTIFY_API_KEY"

Codex sends the Bearer header using that environment variable. See the official Codex MCP configuration.

Cursor

Open Cursor Settings → MCP to add a remote server, or use the URL-and-header JSON above in .cursor/mcp.json. Replace the key placeholder and keep that file out of version control.

VS Code

Run MCP: Add Server and select HTTP, or put this in .vscode/mcp.json. The password input prompts for your key so it is not stored in the shared configuration:

{
  "inputs": [{
    "id": "promptify-key",
    "type": "promptString",
    "description": "Promptify API key",
    "password": true
  }],
  "servers": {
    "promptify": {
      "type": "http",
      "url": "https://www.promptify.pro/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:promptify-key}"
      }
    }
  }
}

Confirm a video render in two steps

1. Ask your assistant to call render_ad({ adId }). It returns the real remaining scene and export quote, including totalCredits. This call spends nothing.
2. Read the quote and approve it. Your assistant then calls render_ad({ adId, confirm: true, expectedCredits }) with expectedCredits equal to that totalCredits. Promptify checks a fresh quote; any change requires approval again.

A confirmed call starts the scene renders through the studio’s paid render service. It returns status and a studio results link; open that link to check progress and finish/export the video. Starting a render does not mean the downloadable video is ready.

Available tools

generate_ad_script

Create an ad with 10 hooks, a scene script and model prompts. Reserves 3 credits, settles to usage and refunds unused credits; failures refund the reservation.

list_ads

List the ads in your workspace.

get_ad

Read an ad’s hooks, script and prompts.

render_ad

Quote a saved ad, then start its paid scene renders after you confirm the exact credit cost.

get_credit_balance

Check subscription, purchased and total remaining credits without spending.

get_pricing_link

Get the credit-pack and plan pricing link.

draft_ad_plan

Estimate a plan’s credit cost before creating an ad. Use render_ad for a saved ad’s fresh quote.

list_templates

Browse ad structures and hook devices.

get_brand_profile

Read your workspace’s saved brand preferences.

scrape_product

Read a product page to ground your script in its details.

get_ad_stats

Read creation, export and real link-click counts. These are not impressions, CTR or ROAS.

Connection and credit help

If OAuth authentication fails, reconnect and approve access again. For API keys, create a key in Settings → API and check the Bearer header. Each credential is tied to its workspace; tool arguments cannot switch workspaces. If credits run out, follow the tool’s pricing link.

Authenticated requests are limited to 60 per token or key per 60 seconds, on a best-effort basis per server instance. A rate-limit rejection spends no credits; wait up to a minute and retry. Initialization, ping and tool discovery work without a key and do not use that budget.