Create an API key in Octoque (You → Developers & AI), then add Octoque’s MCP server to your assistant with that key. After that you can type “schedule my new video to Instagram and TikTok at golden hour” and the assistant does it through Octoque, with the same limits and approvals as the app.
Today the server runs on your own computer, which works with Claude Code and Claude Desktop. ChatGPT and claude.ai’s custom connectors need a hosted web address, which we have not switched on yet.
- MCP (Model Context Protocol) is “an open-source standard for connecting AI applications to external systems”.[1]
- It has two standard transports: stdio (the assistant starts the server on your computer) and Streamable HTTP (the server lives at a web address).[2]
- Claude Code adds a local server with
claude mcp add … -- <command>; everything after--is the command that starts the server.[3] - ChatGPT connects to MCP servers through “a public HTTPS endpoint” (or a secure tunnel), not a program on your computer.[5]
- Octoque’s server has 9 tools and only uses Octoque’s public API, so your plan limits, key scopes and approvals still apply.
What MCP is, in one paragraph
MCP is a shared plug for AI assistants. The MCP project describes it as “like a USB-C port for AI applications”[1]: an app publishes a list of tools (“list my accounts”, “schedule a post”), and any assistant that speaks MCP can call them when you ask. Octoque publishes its tools this way, so you don’t need a special plug-in per assistant.
What you need first
- An Octoque account with at least one platform connected. Using the API and AI assistants is part of every paid plan (see pricing).
- An API key: in the app, open You → Developers & AI → New API key. Name it after the assistant (“Claude”), keep the default permissions (read posts, create drafts, schedule and publish) and copy the key. It starts with
oq_live_and is shown only once. - The start command for the MCP server, which the same page shows. It is a small Python program from Octoque.
Connect Claude Code (one command)
In a terminal, add the server with your key. Claude Code’s docs explain the two traps: options like --env go before the server’s name, with at least one other option between --env and the name, and the -- separates Claude’s options from the server’s own command.[3]
claude mcp add --env OCTOQUE_API_KEY=oq_live_… --transport stdio octoque -- \
<path to Octoque's python> -m scheduler.mcp_serverReplace the last line with the start command from Developers & AI. Add --scope user if you want Octoque in every project rather than only the current folder.[3] Then start Claude Code and type /mcp to check that “octoque” is connected.
Connect Claude Desktop
- Open Claude Desktop’s settings from the menu bar, go to Developer and choose Edit Config. This opens (or creates)
claude_desktop_config.json: on a Mac it lives in~/Library/Application Support/Claude/, on Windows in%APPDATA%\Claude\.[4] - Add Octoque under
mcpServers:{ "mcpServers": { "octoque": { "command": "<path to Octoque's python>", "args": ["-m", "scheduler.mcp_server"], "env": { "OCTOQUE_API_KEY": "oq_live_…" } } } } - Quit Claude Desktop completely and open it again.[4] Octoque now appears in the connectors list in a new chat.
ChatGPT and claude.ai custom connectors
Both of these connect from the provider’s cloud, so they need the MCP server at a public web address. ChatGPT’s developer docs: open ChatGPT’s plugins page, choose Add custom MCP server, and enter the URL “including the /mcp path”; it supports “SSE and streaming HTTP”.[5] Claude’s custom connectors work the same way (Customize → Connectors → Add custom connector), and the server “must be reachable over the public internet”.[6]
Octoque’s server can already speak Streamable HTTP, but only on your own computer for now, because anyone who reaches that address acts with your key. A hosted, signed-in address is planned. We will add the exact steps here the day it goes live. Until then, use Claude Code or Claude Desktop. Which ChatGPT plans can add custom MCP servers changes often; check OpenAI’s help centre for your plan.
The tools your assistant gets
These are the real tool names from Octoque’s server. The assistant picks them for you; you never have to type them.
| Tool | What it does |
|---|---|
list_accounts | Your connected accounts: platform, handle and whether each one is healthy. |
best_times | The next golden hour per platform, in your audience's time zone. |
schedule_post | Schedules a post: at golden hour (default), in the next queue slot, now, or at an exact time with a time zone. Takes per-platform captions, a YouTube title and a local video or image. |
save_draft | Saves a draft without scheduling, so you finish it in the app. |
list_posts | Recent posts, filtered by status (draft, scheduled, published, failed, partial) or a search word. |
get_post | One post with each platform's status, link and any problem. |
retry_post | Retries the platforms that failed. |
cancel_post | Unschedules a post. Anything already live stays live. |
post_results | Views, likes, comments and shares per platform. |
Before uploading, schedule_post checks each caption against the platform’s length (for example 280 characters on X, 2,200 on Instagram and TikTok) and tells the assistant to shorten it rather than letting the platform refuse it later. Videos must be MP4 or MOV; images JPG, PNG or WebP. TikTok needs you to choose who can see each video, and the assistant can’t make that choice for you yet, so TikTok posts are refused with “Choose who can see this on TikTok”: finish those in the app.
Prompts that work
- “What accounts do I have connected on Octoque?”
- “Schedule ~/Videos/edit-tips.mp4 to Instagram and YouTube at golden hour. Caption: 3 editing mistakes I stopped making. YouTube title: 3 Editing Mistakes to Skip.”
- “Write a 250-character version of this caption for X and schedule everything for Friday 7 pm Dubai time.”
- “Save this as a draft for LinkedIn only, I’ll check it in the app.”
- “Did anything fail this week? Retry it if the problem was temporary.”
- “How did last Tuesday’s video do on each platform?”
Keeping it safe
- The server tells the assistant to “never post without the user’s clear go-ahead” and to prefer a draft when unsure. ChatGPT also treats tools that change things as write actions that need your confirmation by default.[7]
- Keys can’t reach account settings: an API key cannot delete your account, change billing or manage your team.
- The key acts as you, with your role. If your posts need a manager’s approval in the app, the assistant’s posts wait for approval too.
- Lost a laptop or shared a key by mistake? Revoke it in Developers & AI; anything using it stops at once.
Questions people ask
Does the assistant see my passwords or platform logins?
No. It only has the Octoque API key you gave it. Your platform connections stay inside Octoque.
Can it post without asking me?
It can if you tell it to and your key allows publishing. The server instructs it to ask first, and you can make a key that can only create drafts.
Does this work with other assistants?
Any app that can run a local MCP server over stdio can use it the same way: give it the start command and the OCTOQUE_API_KEY setting.
Is there a plain API too?
Yes. The same key works with Octoque’s REST API, which the MCP server itself uses. See developers.
Why does the post go out at a different time on each platform?
Because golden hour is per platform. Say “at 7 pm” if you want one time everywhere. What a golden hour is.

Kumar Gaurav builds Octoque. Creators should spend their time making things, not watching clocks. So you make the video, and Octoque finds the moment: one upload, every platform, each at its golden hour. These guides follow the same idea. Real channels, real numbers, nothing you don’t need.
More from Kumar GauravKeep reading
Sources
All links checked on 14 October 2026.
- Model Context Protocol: What is MCP?
- MCP specification: Transports
- Claude Code docs: Connect Claude Code to tools via MCP
- Model Context Protocol: Connect to local MCP servers (Claude Desktop)
- OpenAI developers: Connect a custom MCP server to ChatGPT
- Claude docs: Custom connectors using remote MCP
- OpenAI developers: MCP and connectors (write actions)