Key takeaways
- SkedCast's MCP server exposes 59 tools (31 read, 28 write) over one OAuth-protected endpoint.
- A conversational agent (Claude, ChatGPT, Cursor) connects via MCP — no code, just a connector URL.
- A custom-built agent (your own LangChain/agent-framework pipeline) connects via the REST API directly.
- Every write action is confirmed, never silent — an agent can't publish, delete, or cancel anything your own role couldn't.
What "AI agent for social media" actually means
In practice it means an LLM-based assistant that can call tools on your behalf — read your content calendar, draft a caption, and (with your confirmation) schedule or publish it — rather than a fully autonomous bot posting unsupervised. The distinction matters: every serious implementation, SkedCast's included, treats publishing as a confirmed action, not a background one.
Two shapes of agent show up in practice: a general-purpose conversational agent you already use (Claude, ChatGPT, an IDE's agent mode) that you connect to your tools via MCP, and a purpose-built agent your own team writes — a Slack bot, a content pipeline, a scheduled job — that calls the REST API directly.
Path 1 — a conversational agent, via MCP
MCP (Model Context Protocol) is the open standard for connecting an AI assistant to external tools and data — Anthropic introduced it and has since donated it to the Agentic AI Foundation, a Linux Foundation project, reflecting how broadly it's been adopted (ChatGPT, Cursor, VS Code/Copilot, and others all speak it today). SkedCast runs a remote MCP server at https://api.skedcast.com/mcp over Streamable HTTP; connect Claude, ChatGPT, Cursor, or another MCP-capable client with just the endpoint URL, and the client runs the OAuth sign-in for you — no code required.
- Claude (web, desktop, or Cowork) — add a custom connector in Settings.
- ChatGPT — enable Developer Mode, then add a custom connector (Pro/Team/Enterprise/Edu).
- Cursor, VS Code, Windsurf — a short JSON entry in each editor's MCP config file.
- Claude Code — one
claude mcp add --transport httpcommand.
Path 2 — a custom-built agent, via the REST API
If you're building your own agent (a LangChain/LangGraph pipeline, a custom orchestrator, a scheduled job that decides what to post), skip MCP and call the REST API directly with an API key or OAuth token — the same compose, media, and webhook endpoints, with full programmatic control over what gets sent and when.
The safety model, either path
An agent's effective power is always your account's scopes ∩ role — connecting Claude or building a custom agent never grants it more than the SkedCast member whose credentials it's using could do by hand. Over MCP specifically, every tool that changes something (create_post, cancel_post, delete_webhook, and the rest) is marked non-read-only, so a well-behaved MCP host prompts you to confirm before it runs. Read tools — listing posts, checking analytics, reading best-times — run freely within your granted scopes.
FAQ
- Can an AI agent publish a post without anyone approving it?
- Only if you've set it up that way on purpose (a custom agent calling the API directly with saveDraft: false and no human in the loop). Over MCP, write tools are flagged non-read-only specifically so a well-behaved host — Claude, ChatGPT, Cursor — asks for confirmation before calling them.
- Do I need to be a developer to connect an AI agent to SkedCast?
- No, for the MCP path — connecting Claude, ChatGPT, or an IDE agent is pasting one URL into that tool's settings, the same as adding any other connector. Building a fully custom agent against the REST API is a developer task.
- Which AI agent should I use for social media?
- Whichever one you already use for everything else — the MCP connection works the same way across Claude, ChatGPT, Cursor, and other MCP-capable clients, so there's no separate "social media AI" product to adopt.