SocialGPT

What is this?

This is the public front door for the SocialGPT MCP server — a hosted, remote MCP server that brings your social media analytics into any MCP-compatible AI agent. There's nothing to install and no API keys to manage: you point your client at one URL and authenticate once with your SocialGPT account.

Connection URL:

https://mcp.gpt.social/mcp

• Transport: Streamable HTTP
• Auth: OAuth 2.1 with PKCE — sign in once with your SocialGPT account, revoke any time
• Access: Read-only analytics by default. Two optional scopes go further — one lets agents pull new content into your library for analysis, the other lets them publish to your connected accounts.

New to SocialGPT? Create a free account, then connect a social account to unlock analysis of your own content. Public/competitor analysis works without connecting anything.

Setup

Pick your client, paste the config, and authenticate with your SocialGPT account. Most clients open a browser window for OAuth the first time an agent uses a SocialGPT tool.

Ready-to-copy config files for each client live in examples/.

Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS), then restart Claude Desktop:

{  "mcpServers": {    "socialgpt": {      "url": "https://mcp.gpt.social/mcp"    }  }}

Claude.ai

Settings → Integrations → Add more → paste the URL and follow the prompts:

https://mcp.gpt.social/mcp

Claude.ai handles authentication during the integration setup flow.

Claude Code

claude mcp add --transport http socialgpt https://mcp.gpt.social/mcp

On your first request that uses SocialGPT tools, Claude Code opens a browser for OAuth sign-in and caches the token locally.

Cursor

Add to your project's .cursor/mcp.json, or ~/.cursor/mcp.json for global access:

{  "mcpServers": {    "socialgpt": {      "url": "https://mcp.gpt.social/mcp"    }  }}

VS Code

Add to .vscode/mcp.json in your workspace (or run MCP: Add Server from the command palette):

{  "servers": {    "socialgpt": {      "type": "http",      "url": "https://mcp.gpt.social/mcp"    }  }}

ChatGPT

Settings → Tools & integrations → Add MCP server, then paste the URL:

https://mcp.gpt.social/mcp

Codex

Add to ~/.codex/config.toml (or .codex/config.toml for a single project):

[mcp_servers.socialgpt]url = "https://mcp.gpt.social/mcp"

Cline

MCP Servers icon → Remote Servers → add the server URL, or edit cline_mcp_settings.json:

{  "mcpServers": {    "socialgpt": {      "type": "streamableHttp",      "url": "https://mcp.gpt.social/mcp"    }  }}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{  "mcpServers": {    "socialgpt": {      "serverUrl": "https://mcp.gpt.social/mcp"    }  }}

AntiGravity

Add to ~/.gemini/config/mcp_config.json:

{  "mcpServers": {    "socialgpt": {      "serverUrl": "https://mcp.gpt.social/mcp"    }  }}

Any other MCP client

Any client that supports the Model Context Protocol can connect using the server URL https://mcp.gpt.social/mcp. Check your client's docs for where to register an MCP server, and use the streamable HTTP transport.

What you can do

Just ask

Once connected, talk to your agent like a teammate:

"How did my content perform this month compared to last?"

"Find creators in the fitness niche on TikTok and show me what's working for them."

"Analyze my latest Instagram reel and suggest better hooks."

"Compare my engagement rate to similar creators on YouTube."

"Post this video to my TikTok as a draft."

Tools

The server exposes a curated surface — read-only unless you grant a write scope. The Scope column shows which OAuth scope each tool needs (see Authentication & scopes).

Your videos & content

Accounts & metrics

Competitor & creator discovery

Analyze new content

Publish

Opt-in — these are the only tools that act on your accounts, and they require the content:publish scope. Nothing publishes without your explicit confirmation.

Publishing a local file is a three-step handoff: get_upload_link → you upload in the browser → list_uploads to find the draft → publish_post(draft_id=...).

Meta

Conventions

A few behaviours that apply across the surface:

• Account groups (brands / workspaces). Accounts are organized into groups. list_accounts returns each account's group plus a groups list; pass a group_id to list_accounts, get_content_profile, get_publish_options, or publish_post to scope to one brand. Omit it for the default group.
• Pagination. list_videos, search_videos, and list_creator_videos return a next_cursor when more results exist — pass it back as cursor with the same filters. No next_cursor means you have everything.
• Time-series granularity. get_account_metrics, get_follower_history, and get_post_metrics_history take a granularity of daily (default), weekly, or raw.
• Async analysis. Deep video analysis runs in the background (~30–60s). get_video_analysis returns {"status": "pending", "retry_after_seconds": N} until it's ready — that's expected, not an error. Wait and call again.

Guided workflows

The server also ships prompts — guided workflows that chain the tools above into a complete analysis. Clients surface these as slash-commands or prompt templates.

Skills

Where prompts ask the model to do an analysis, skills ship the work itself. Most are a real script the agent runs on your MCP data, returning a styled report; going-viral is the strategy loop that conducts the others. They install on Claude.ai (upload a zip) and Claude Code (drop in a folder), and run the same SocialGPT MCP tools under the hood.

See skills/ for install instructions, the build script, and the methodology behind each one. New skills land here as the tool surface grows.

Authentication & scopes

The SocialGPT MCP server is an OAuth 2.1 resource server. Your client performs an OAuth 2.1 + PKCE flow against your SocialGPT account; the server validates the bearer token on every request. No API keys are ever copied around, and you can revoke access at any time from your SocialGPT settings.

Access is governed by four scopes. The two read scopes are granted by default; the two write scopes are opt-in — a client has to ask for them explicitly, and you have to approve them.

Clients can also request offline_access for a refresh token, so you don't re-authenticate every hour.

Until you connect a social account, only public/competitor analysis is available. Newly connected accounts appear automatically on the same token — no need to re-authenticate.

Roadmap

Read access and publishing are live today. We're building the rest of the write surface so your agent can do more than analyze and post — all over the same MCP connection:

• Schedule posts — queue content for optimal posting times
• Generate scripts — full video scripts grounded in your analytics and style
• Design carousels — generate carousel posts from your content DNA (publishing an existing carousel already works)
• Manage comments — read, reply to, and moderate comments across platforms
• A/B test hooks — run experiments on hooks and captions

Support & links

⭐ If SocialGPT MCP is useful to you, star this repo — it helps other creators discover it.

• Product: gpt.social · MCP integration page
• Sign up: app.gpt.social/signup
• Connect accounts: app.gpt.social/account
• Learn about MCP: modelcontextprotocol.io

This repository contains the public listing and connection details for the SocialGPT MCP server. The server itself is hosted by SocialGPT; its source is not part of this repository.

License

MIT © Scrollmark, Inc.