The Envox MCP server lets the AI assistant you already use work in Envox for you:
- Publish — the assistant writes the caption and brings images or a video; Envox posts them now, at a time you choose, or keeps a draft for you to approve.
- Create — on-brand images, carousels and Reel videos with Envox AI, in the style of your brand guide.
- Plan and measure — post ideas from your best posts and your competitors, and the likes, reach and views of what you published.
- WhatsApp Business — write and submit templates, send them to your customers, read and answer conversations.
You pay only for what Envox does: a small credit per account for publishing your own media, credits per AI image or video (published at no extra cost), and per WhatsApp message. Statistics, template drafts and reading are free.
Before you start
- At least one connected account in Social Connections: an Instagram business account, a Facebook page or a LinkedIn profile. For WhatsApp, a number in Phone Numbers; for AI images, a brand guide.
- An AI client that can send a custom header: Claude Desktop, Claude Code, Cursor, VS Code or Windsurf.
- Credits on the account — see Credits. The assistant can tell
you every price (it is in
get_account).
ChatGPT and the claude.ai web connectors ask for an OAuth sign-in, which is coming soon. Until then, use one of the clients above.
Steps
- 1
Create an API key
Open API & MCP in the panel menu and click New key. Give it a name you will recognise later — for example Claude Desktop — and click Create.
The key starts with
envx_and is shown only once. Copy it before closing the window. Right after you create it, the configurations further down the page already contain the key, ready to copy.The key acts as you. Do not paste it into a shared document or commit it to a repository. If it leaks, revoke it on the same page and create a new one.
- 2
Add Envox to your assistant
The server address is:
https://app.envox.ai/mcpIt uses Streamable HTTP and expects the header
Authorization: Bearer envx_…. Pick your client:Claude Code — run in a terminal:
claude mcp add --transport http envox https://app.envox.ai/mcp \ --header "Authorization: Bearer envx_YOUR_API_KEY"Claude Desktop — Settings → Developer → Edit config, add the server to
claude_desktop_config.json, save and restart Claude. Node.js must be installed.{ "mcpServers": { "envox": { "command": "npx", "args": ["-y", "mcp-remote", "https://app.envox.ai/mcp", "--header", "Authorization:${ENVOX_AUTH}"], "env": { "ENVOX_AUTH": "Bearer envx_YOUR_API_KEY" } } } }Cursor —
~/.cursor/mcp.json:{ "mcpServers": { "envox": { "url": "https://app.envox.ai/mcp", "headers": { "Authorization": "Bearer envx_YOUR_API_KEY" } } } }VS Code —
.vscode/mcp.jsonin your project:{ "servers": { "envox": { "type": "http", "url": "https://app.envox.ai/mcp", "headers": { "Authorization": "Bearer envx_YOUR_API_KEY" } } } } - 3
Ask for what you need
Talk to the assistant the way you would brief a colleague:
- “Post this image on Instagram and our Facebook page now, with a short caption.”
- “Schedule the five designs in this folder, one per day at 10:00 next week.”
- “Post reel.mp4 from my desktop as a Reel tomorrow at 19:00 and write a short caption for it.”
- “Make a Black Friday image for Instagram in our brand style, then turn it into a short Reel.”
- “Give me five post ideas for next week based on what worked for us and for @rivalcafe.”
- “How did our Instagram posts do this month? Which one was best and why?”
- “Write a WhatsApp template that reminds customers of tomorrow's appointment and send it to these three numbers once it is approved.”
- “Are there WhatsApp messages I haven't answered today?”
Before publishing, sending a WhatsApp message or replying to a customer, the assistant shows you what will go out and waits for your yes.
Times without a timezone, such as “tomorrow at 9”, are read in the timezone set in Settings. Set it once and every schedule lands where you expect.
- 4
Check the result
Tools
The assistant picks these on its own; you never call them by name.
Posts
| Tool | What it does |
|---|---|
list_social_accounts |
The connected accounts and their ids |
create_post |
A post from images or a video and a caption — now, scheduled or as a draft |
publish_post |
Publishes or schedules a draft, also one made in the panel |
upload_media |
Copies an image or a video from a public URL into Envox |
create_media_upload_link |
A one-time link to upload a file from your computer |
list_posts / get_post |
Posts, and the result on each account |
update_post |
Changes the caption or moves a scheduled post |
cancel_post |
Turns a scheduled post back into a draft |
delete_post |
Deletes a draft |
get_account |
Timezone, credit balance and prices |
Envox AI
| Tool | What it does |
|---|---|
generate_image |
On-brand images (up to 4 variations) or a planned carousel from a topic |
edit_image |
A changed copy of an image |
generate_video |
Turns an image into a short motion video for Reels |
get_generation |
Progress and result of a generation |
list_brand_guides |
Brand guides and reference image sets |
Ideas and performance
| Tool | What it does |
|---|---|
generate_suggestions / get_suggestion_batch |
New post ideas from your and your competitors' best posts |
list_suggestions / dismiss_suggestion |
The ideas, and dropping the ones you don't want |
list_competitors / add_competitor / remove_competitor |
Competitor Instagram accounts the ideas learn from |
get_post_insights |
Likes, comments, reach, views and saves of a post on each account |
get_account_insights |
An account's last 30 days: totals, every post, the best post |
WhatsApp Business
| Tool | What it does |
|---|---|
list_whatsapp_numbers |
The connected WhatsApp numbers |
list_whatsapp_templates |
Templates and their Meta review status |
draft_whatsapp_template |
Envox AI writes a template that follows Meta's rules |
create_whatsapp_template |
Submits a template to Meta for review |
send_whatsapp_template / get_whatsapp_send |
Sends a template to one or many numbers, and the delivery report |
list_whatsapp_conversations / get_whatsapp_conversation |
Customer conversations and their messages |
reply_whatsapp_conversation / release_whatsapp_conversation |
Answers a customer (within 24 hours) and hands back to the AI agent |
Good to know
- Images: JPEG, PNG or WebP up to 8 MB, 1 to 10 per post. PNG and WebP are converted to JPEG, which Instagram requires. For Instagram feed posts keep the aspect ratio between 4:5 and 1.91:1.
- Videos: one MP4 or MOV per post, up to 100 MB and 3 seconds to 15 minutes, ideally 9:16. It becomes a Reel on Instagram and a video on Facebook pages; LinkedIn does not take videos yet.
- Stories: one image or one video up to 60 seconds, Instagram only.
- Files on your computer: clients with terminal access (Claude Code, Cursor) upload them through a one-time link that expires after 15 minutes. Other clients need a public file link.
- Captions: up to 2,200 characters for Instagram.
- AI images and videos: made in the style of your brand guide; images take 20–90 seconds, videos 1–4 minutes, and the assistant waits for them. At most five run at the same time.
- Statistics: reach, views and saves need Instagram's insights permission; accounts connected before Envox asked for it show likes and comments until you reconnect them.
- WhatsApp: templates go out only after Meta approves them (minutes to a day). Free-form replies are possible within 24 hours of the customer's last message. Replying takes the conversation over from your AI agent until the assistant hands it back. Only message people who agreed to hear from you.
- Credits: your own media is charged per account when published; AI images, slides and videos when generated (refunded if they fail) and they publish at no extra cost; WhatsApp templates per sent message. Drafts, statistics, template drafts and reading are free. When credits run out the assistant tells you instead of retrying.
- Team members: a key works with the permissions of the user who created it — content permissions for posts and AI, phone number permissions for WhatsApp templates, inbox permissions for sending and replying. See Roles.
- Revoking: a revoked key stops working immediately for every client using it.
- For your own code instead of an assistant, see the API reference.