Practical setup guides, publishing workflows, and a closer look at the protocol connecting your AI assistant to every social channel.
Setup guide · 10 min read
Every screen, setting, and config snippet in order: sign in to PostMCP, connect your social profiles, copy your remote MCP server URL, add the custom connector in Claude.ai or Claude Desktop, and publish your first live post directly from chat.
Setup guide · 10 min read
Every screen, toggle, and setting in order: sign in to PostMCP, connect your social profiles, enable Developer Mode in ChatGPT, add the remote MCP connector, and publish your first live post directly from chat.
What the Model Context Protocol is, and what it changes for publishing.
There are dozens of social MCP servers on npm and most of them wrap a single post-a-tweet call. Here is the checklist that separates a weekend demo from something you can leave running unattended.
Wiring a specific client to a specific network, step by step.
Every click and screen in order: open your profile menu, navigate to Settings, select Advanced, toggle Developer Mode ON, and unlock custom Model Context Protocol (MCP) connectors.
Every screen you will actually see, in order: the sign-in page, the channel connection, the MCP server URL, Claude’s custom connector dialog, and the first post going out. Roughly five minutes end to end, most of it waiting on OAuth.
Five lines of JSON, one API key, and Claude can draft, preflight and schedule LinkedIn posts without you opening a browser tab. Here is the exact setup — plus the four things that trip up the first attempt.
ChatGPT can publish to seven networks two different ways, and the choice is not about capability — it is about who you are building for and who holds the key.
X is the network where AI-written copy fails most visibly: everything is thirty characters too long and reads like a summary. What to do about it, mechanically.
Running a real posting workflow with an AI agent in the loop.
The phrase is used for everything from a caption generator to a fully autonomous account. Here is what a real AI social media manager does, where the line with a human sits, and what to check before you trust one with your accounts.
The failure mode of AI social automation is not bad posts. It is a feed that never stops and never says anything. Here is the split between what an agent should own and what it should never touch.
The model never holds your credentials — that part is genuinely solved. The risks that remain are approval fatigue, prompt injection and key sprawl, and none of them are fixed by architecture alone.
Building on the protocol: architecture, transports, failure modes.
Both can publish the same post to the same network. The difference is who decides what to send — your code, or a model reading a tool schema. That single distinction determines everything else.
The MCP part takes an afternoon. Everything that makes it survive contact with seven social platforms takes considerably longer. Notes from building one, including the mistakes.
Seventeen publishing and analytics tools across seven networks, MIT licensed on npm. The free plan covers 20 posts a month and never expires.
npx -y @postmcpai/server