Where do I create an MCP key?

In Postemi, on the Automations → Developers page. The MCP key starts with psty_mcp_, and that is what you put in your MCP client setup. We only show the key when you create it, so save it right away.

How do I install the command-line tool?

With npm install -g postemi. The npm package and the installed command are both named postemi. The source is at github.com/norbertlevente/postemi-agent, licensed AGPL-3.0.

Can the agent upload images and video too?

Yes. First it uploads the file with postemi upload, gets a URL back, and passes that to the -m flag. The publishing flow rejects a plain filename and an external link too.

Which plan includes all of this?

The MCP server, the command-line tool and the public REST API are part of the Pro, Creator and Team plans. There is no extra charge, and you do not have to negotiate it with anyone.

Does Postemi write the post for me?

No, and that is on purpose. Postemi has no built-in AI writer: the agent you already work with writes the text. Postemi handles the connection, the scheduling, the publishing and the calendar.

Post with AI agents: ChatGPT, Claude and MCP

Getting started

Four commands, and it runs

You install, sign in, list the connected accounts, then schedule the first post. Your agent runs the same commands.

terminal

# The Postemi command-line tool
$ npm install -g postemi
# Sign in: approve in the browser
$ postemi auth:login
# Which social accounts are connected?
$ postemi integrations:list
[{ "id": "cmf3k7…", "name": "Cafe · Instagram" }]
# The first scheduled post
$ postemi posts:create \
$   -c "New brunch menu this Saturday." \
$   -i "cmf3k7…" \
$   -s "2026-08-16 09:00" --timezone Europe/London
✅ Post created successfully!
{ "id": "cmf7p2…", "state": "QUEUE" }

Needs the Pro plan or higher. Every plan starts with a 7-day free trial. The CLI writes the JSON response to stdout, and status and errors to stderr. On failure it exits with a non-zero code. That way the agent knows whether it worked.

What you get

Three ways in, one Postemi account

Want to chat? MCP. Want to script? CLI. Want to integrate? REST API. All three work today.

Live MCP server

Your agent connects to https://mcp.postemi.com/mcp, and from then on it sees the same Postemi account you do. You create the key on the Automations → Developers page.

Published command-line tool

Install with: npm install -g postemi. The command name is postemi.

Public REST API

The REST API is at api.postemi.com. It handles the same account as MCP and the CLI, just from your own code.

No training needed

SKILL.md ships in the npm package. The agent reads the real commands and flags from it, so you do not have to explain them.

The calendar is where you approve

The agent prepares and schedules. You have the last word before anything publishes. In the calendar you edit, reschedule or delete what you do not want.

You keep your own model

Postemi has no built-in AI writer, on purpose. You work with the model you already use. Postemi handles the publishing.

How it works

From accounts to approval

You set up the connection once. After that your agent does the work, and you decide in the calendar.

01

Connect your accounts

On the Postemi web app, you connect your social accounts with the usual OAuth flow. The agent later works with those accounts and refers to them by id.

Ids of the connected accounts

terminal

$ postemi integrations:list
[
  { "id": "cmf3k7…", "name": "Cafe · Instagram" },
  { "id": "cmf3k9…", "name": "Cafe · Facebook" }
]

02

Create a key

Automations → Developers. Chat clients need an MCP key. The CLI and the REST API need an API key. We show the key once, when you create it.

Automations → Developers

.env

# Automations → Developers → New key
# We show the key once, when you create it.
 
# For the command line and the REST API:
POSTEMI_API_KEY=YOUR_API_KEY
 
# MCP clients get a separate key:
# psty_mcp_…

03

Hand it to the agent

Chat clients only need a URL. Terminal clients need a command or a config block. On the Developers page you pick the client and copy the ready-made setup.

Ready-made setup for every client

MCP setup

# Browser assistant (ChatGPT, Claude app):
# add a custom connector with this address, then sign in:
https://mcp.postemi.com/mcp
 
# Claude Code (signs in through the browser):
claude mcp add --transport http postemi https://mcp.postemi.com/mcp
 
# Codex, ~/.codex/config.toml, then: codex mcp login postemi
[mcp_servers.postemi]
url = "https://mcp.postemi.com/mcp"

04

Review it, then it can go

The agent puts the posts in the calendar. You go through them, fix what needs fixing, and only leave what can actually go out.

Scheduled, but not published yet

terminal

$ postemi posts:list
[
  { "id": "cmf7p2…", "state": "QUEUE",
    "publishDate": "2026-08-16T07:00:00Z" },
  { "id": "cmf7q8…", "state": "DRAFT",
    "publishDate": "2026-08-18T06:30:00Z" }
]

Built for agents

Predictable input, a clear response

The agent does not have to fuss with the web UI. It gets a structured response, and it reaches the same account three ways.

Machine-readable JSONEvery command writes the JSON response to stdout, so the agent can parse it directly.
An error is really an errorThe CLI writes status and errors to stderr. On failure it exits with a non-zero code.
Skill file in the packageSKILL.md shows the available commands, the flags and the two rules you cannot skip.
A flat command setNo hidden state, and no step to guess: the agent works from a clear list.
Three ways to the same placeMCP for the conversation, CLI for scripts, REST API for your own code: the same account.

SKILL.md

---
name: postemi
description: Postemi schedules and publishes social posts
  from the command line.
homepage: https://postemi.com
---
 
# Output contract
 
Every command prints its JSON result, and nothing
else, to stdout; status lines, warnings and errors
go to stderr, and every failure exits 1.
 
# Two Hard Rules (Read First)
 
Rule 1: Authenticate before anything.
Rule 2: Every file passed to `-m` MUST first go
through `postemi upload`.

SKILL.md is part of the npm package. The agent reads the exact CLI usage from it: no extra training, and no invented flags.

Supported clients

The same Postemi, several agents

Each one has its own page with the exact setup. If yours is not among them, that is fine: Postemi uses a standard MCP server.

ChatGPT

Add Postemi to ChatGPT as a custom MCP server, with a browser sign-in.

Your agent schedules the posts for you

Four commands, and it runs

Three ways in, one Postemi account

Live MCP server

Published command-line tool

Public REST API

No training needed

The calendar is where you approve

You keep your own model

From accounts to approval

Connect your accounts

Create a key

Hand it to the agent

Review it, then it can go

Predictable input, a clear response

The same Postemi, several agents

ChatGPT

OpenAI dots

Claude

Claude Cowork

Claude Code

Codex

Cursor

Gemini

OpenClaw

Hermes Agent

Grok Bot

Grok Build

DeepSeek

Kimi K3

Muse

Manus

Perplexity Computer

NanoClaw

Paperclip

MCP Server

AI Agents CLI

What your agent can run

One account, every channel

TikTok

Instagram

Facebook

Threads

LinkedIn

YouTube

X

Pinterest

Bluesky

Google Business

Discord

Telegram

Slack

Mastodon

Snapchat

Pixelfed

Tumblr

WordPress

Ghost

Dev.to

Google Chat

Microsoft Teams

Hashnode

Lemmy

Nostr

Dribbble

Twitch

Kick

Whop

Webflow

Shopify

Circle

listmonk

MailerLite

MailCheap

Mailchimp

Kit

Brevo

Buttondown

Fair pricing, no hidden costs

Basic