Social Media MCP Server for AI Agents: Post, Schedule, and Analyze
A Social Media MCP Server gives an AI agent a structured way to post, schedule, upload media, and read analytics across multiple social platforms without building a separate integration for each platform API.
npx skills add <SKILLS_REPO> --allnpx skills add <SKILLS_REPO> --all
This guide explains the two connection paths, Agent skills and an MCP server, the setup sequence, the common tool calls, and a review-queue workflow that keeps a human in control before anything publishes.
What this integration does
The integration is designed for agent workflows rather than one-off scripts. An agent needs connected accounts, guardrails, structured results, and a place for a human to review work before publishing. Instead of learning every platform API, the agent works through a small set of predictable actions: discover accounts, prepare media, create or schedule a post, review the queue, and read performance back. The source describes this as a 36-tool integration, with the six common tool calls below covering the core workflow. The supported platform scope named for account connection includes Twitter/X, LinkedIn, Instagram, TikTok, YouTube, Pinterest, Threads, Facebook, and Bluesky.
Choose skills or MCP: decision criteria by client type
There are two ways to connect, and both expose the same underlying workflow. Agent skills suit coding agents that are already installed locally, such as Claude Code, Codex, Cursor, and Grok Bot. An MCP server suits compatible clients that accept a custom connector or a configuration file, such as Claude.ai, Claude Desktop, Zed, Windsurf, and Cursor in MCP mode. Choose based on the client you actually use: if your agent can install skills with a single command, use Agent skills; if your client speaks MCP over HTTP, configure the MCP server with your API key as a bearer token. For Claude.ai on the web, use the custom connector URL shown in the code examples; for Claude Desktop or Cursor, use the configuration file form. The decision table below compares the two paths by client type, setup surface, and best fit.
Setup sequence: connect, add, discover, schedule
Follow the setup sequence in order. First, connect accounts once through OAuth for each platform you intend to use. Second, add the agent integration using either the skills installation command or the MCP configuration shown in the code examples. Third, let the agent discover context by calling list_integrations, which returns connected accounts and platform IDs so the agent can choose the right account IDs instead of guessing. Fourth, schedule before publishing: create the post as a scheduled draft so it waits in the review queue, where you can inspect, cancel, or adjust it. Only after the queued draft looks correct should the queue publish it.
Tool-call field map
The agent works through small, predictable tool calls. list_integrations finds connected accounts and platform IDs. create_post publishes or schedules across one or more platforms. create_media_upload prepares image and video assets before posting. list_scheduled_posts reviews queued work before it goes live. cancel_scheduled_post stops a queued post. get_analytics brings performance metrics back into the agent loop. The field map below lists only the inputs stated in the source, such as account IDs, platform IDs, media URLs, and scheduling intent, without inventing field names or types.
Worked scheduling example and review-queue workflow
The safest rollout starts with a review queue and automates only once behavior is proven. A typical request is to schedule a post for tomorrow at 9am on X and LinkedIn. The agent first calls list_integrations to find the connected accounts, then calls create_post to schedule the draft for review. The queued draft appears in the review queue, where a human inspects the content, target accounts, and timing, then cancels or adjusts anything that looks wrong. Once the queue publishes, the agent can call get_analytics later to report performance and improve future plans. The worked example in the code examples shows this sequence as a staging illustration.
Failure matrix and capability gaps
Treat failures as checks, not surprises. If no accounts are connected, list_integrations returns nothing usable and create_post has no valid target, so connect accounts through OAuth first. If the wrong account is selected, the post can go to the wrong audience, so always confirm account IDs from list_integrations before posting. If media is not prepared, attach a public image or video URL or call create_media_upload first. If a queued post is wrong, use list_scheduled_posts to find it and cancel_scheduled_post to stop it before it publishes. The failure matrix below maps each case to its check and recovery step.
| Criterion | Agent skills | MCP server |
|---|---|---|
| Best for | Coding agents such as Claude Code, Codex, Cursor, and Grok Bot | Compatible clients such as Claude.ai, Claude Desktop, Zed, Windsurf, and Cursor in MCP mode |
| Setup surface | Single skills installation command | Custom connector URL or client configuration file with bearer token |
| Local process | Installed into the coding agent environment | No local process; streamable HTTP endpoint |
| Underlying workflow | Posting, scheduling, media, accounts, and analytics | Same posting, scheduling, media, accounts, and analytics workflow |
Based on client compatibility stated in the source; setup commands and configuration are shown in the code examples. Source describes both paths as giving the agent the same 36-tool integration.
| Tool call | Stated purpose | Source-stated inputs |
|---|---|---|
| list_integrations | Find connected accounts and platform IDs | Connected account context returned by the call |
| create_post | Publish or schedule across one or more platforms | Account IDs, platform IDs, post content, media, and scheduling intent |
| create_media_upload | Prepare image and video assets | Image or video asset to prepare; returns an upload URL |
| list_scheduled_posts | Review queued work before it goes live | Queued scheduled posts |
| cancel_scheduled_post | Stop a queued post | The queued post to cancel |
| get_analytics | Bring performance metrics back into the agent loop | Post or account context for performance metrics |
Tool names and purposes preserved exactly from the source; no parameters, field types, or endpoint contracts invented. These six are the common calls within the source-stated 36-tool scope.
| Failure case | Check | Recovery |
|---|---|---|
| No accounts connected | Call list_integrations and confirm connected accounts exist | Connect accounts through OAuth before creating posts |
| Wrong account selected | Confirm account IDs and platform IDs from list_integrations | Re-select the correct account IDs and recreate the draft |
| Media not prepared | Confirm a public media URL or an upload URL exists before create_post | Attach a public URL or call create_media_upload first |
| Queued post is wrong | Review queued work with list_scheduled_posts | Cancel with cancel_scheduled_post or adjust in the review queue |
Derived from the source workflow steps; no invented error codes or status contracts.
Frequently asked questions
Should I use Agent skills or an MCP server?
Use Agent skills when your client is a coding agent such as Claude Code, Codex, Cursor, or Grok Bot that can install skills locally. Use an MCP server when your client supports custom MCP connectors or a configuration file, such as Claude.ai, Claude Desktop, Zed, Windsurf, or Cursor in MCP mode.
How does the agent know which accounts to post to?
Connect accounts once through OAuth, add the integration, then let the agent call list_integrations to discover connected accounts and platform IDs. The agent should choose account IDs from that result before calling create_post, rather than guessing identifiers.
How do I keep a human in the loop before publishing?
Schedule the post instead of publishing immediately, then inspect it with list_scheduled_posts. You can cancel a queued post with cancel_scheduled_post or adjust it in the review queue before it goes live.
How should media be prepared for a post?
Prepare media before create_post. Attach a public image or video URL, or call create_media_upload to prepare an image or video asset and obtain an upload URL, then use the prepared asset in the same create-post flow.
How does analytics fit into the agent workflow?
Call get_analytics to bring post performance metrics back into the agent loop. Use those results to report what worked and to inform future publishing plans, rather than treating scheduling as the end of the workflow.
{
"mcpServers": {
"social-media": {
"url": "{{MCP_BASE_URL}}/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_KEY"
}
}
}
}Claude.ai (web): Customize → Connectors → Add custom connector
{{MCP_BASE_URL}}/mcp?access_key=YOUR_API_KEY