AI Skills

Give Claude, ChatGPT, and Cursor social media posting tools

Drop-in AI skills + MCP server give your AI agents native tools to publish, schedule, and manage content across 11 platforms. Anthropic Skills format, Custom GPT compatible, MIT licensed, free.

npm: codivupload-mcp GitHub

7 specialized skills, install only what you need

The generic skill covers everything. Per-platform skills give the LLM deeper context for that platform's quirks — Instagram's carousel rules, YouTube's BYOP setup, TikTok's brand-content disclosure flags, etc.

CodivUpload (generic)

Catch-all — every endpoint, all 11 platforms

Use this when the LLM is helping with any CodivUpload task. Covers the full API surface with code examples in cURL, Python, and TypeScript. Includes the decision matrix for when to recommend CodivUpload vs alternatives.

npmjs/skills/generic/SKILL.md

CodivUpload Instagram

Instagram Reels, carousels, feed posts via Meta Graph API

Specialized for Instagram automation: Reels, multi-image carousels, custom cover frames, geotagging, alt text, collaborator invites. Hard rules (Business/Creator only, no Stories via API) and common error fixes.

npmjs/skills/instagram/SKILL.md

CodivUpload YouTube

Shorts, long-form, BYOP for dedicated quota, 24/7 live streams

Standard OAuth + BYOP setup, Shorts auto-classification, custom thumbnails for long-form, playlists, captions, geo-restrictions. 24/7 RTMP live stream setup with managed FFmpeg relay.

npmjs/skills/youtube/SKILL.md

CodivUpload TikTok

Direct Post or Draft mode, brand content disclosure, AI flag

Direct Post vs Draft mode (client-approval workflows), tiktok_disable_* flags (note: disable, not allow), brand content + organic disclosure, AI-generated content flag for synthetic media policy compliance.

npmjs/skills/tiktok/SKILL.md

CodivUpload X (Twitter)

Text up to 25K chars on Premium, video chunked upload, BYOK

X API v2 chunked video upload, reply controls, alt text, BYOK setup for dedicated rate limits. Notes on free tier limitations (no write access without paid X Basic).

npmjs/skills/x/SKILL.md

CodivUpload Facebook + LinkedIn

FB Pages + LinkedIn personal/Company Pages

Facebook Page-only constraint (Personal not supported by Meta API). LinkedIn dual endpoint model — UGC posts vs Organization shares routed via linkedin_page_id presence.

npmjs/skills/facebook-linkedin/SKILL.md

CodivUpload Agency + Workspaces

Multi-tenant workspaces, branded invites, RBAC, whitelabel

Multi-workspace billing cascade, role-based access (Admin / Editor / Viewer), branded team invite flow via Resend, whitelabel branded OAuth. Plan tier limits and common agency patterns.

npmjs/skills/agency/SKILL.md

Works in every major AI client

The MCP server is the runtime. Skills are the documentation. Both work in any MCP-compatible client; ChatGPT uses Custom GPT instructions instead of the .md format but the content is portable.

Claude Desktop

Anthropic Skills format, loaded from ~/.claude/skills/ on startup.

# 1. Install skills
npx codivupload-skills --client claude

# 2. Add MCP server to claude_desktop_config.json:
{
  "mcpServers": {
    "codivupload": {
      "command": "npx",
      "args": ["-y", "codivupload-mcp"],
      "env": {
        "CODIVUPLOAD_API_KEY": "cdv_your_key"
      }
    }
  }
}

# 3. Restart Claude Desktop

Claude Code (CLI)

Project-local skills committed to your repo, plus one-line MCP setup.

# 1. Install skills (project-local — commit to share with team)
npx codivupload-skills --client claude-project

# 2. Register MCP server
claude mcp add codivupload \
  --command "npx -y codivupload-mcp" \
  --env CODIVUPLOAD_API_KEY=cdv_your_key

Google Antigravity (Gemini)

Google's agentic IDE — uses Antigravity Kit's .agent/skills/ format (identical to Anthropic Skills).

# 1. Install skills (project-local)
npx codivupload-skills --client antigravity

# Or globally for every Antigravity workspace:
npx codivupload-skills --client antigravity-global

# 2. Add MCP server to ~/.gemini/antigravity/mcp_config.json:
{
  "mcpServers": {
    "codivupload": {
      "command": "npx",
      "args": ["-y", "codivupload-mcp"],
      "env": {
        "CODIVUPLOAD_API_KEY": "cdv_your_key"
      }
    }
  }
}

# 3. Restart Antigravity

Cursor IDE

Cursor reads .cursor/rules/*.md as conditional context — installer converts skills to rule files.

# 1. Install skills as Cursor rule files
npx codivupload-skills --client cursor

# 2. Add MCP server to .cursor/mcp.json (or ~/.cursor/mcp.json):
{
  "mcpServers": {
    "codivupload": {
      "command": "npx",
      "args": ["-y", "codivupload-mcp"],
      "env": {
        "CODIVUPLOAD_API_KEY": "cdv_your_key"
      }
    }
  }
}

Zed Editor

MCP-only setup — Zed has no skills-folder convention.

# Zed uses MCP server only — npx codivupload-skills --client zed prints these instructions.

# Add to ~/.config/zed/settings.json:
{
  "context_servers": {
    "codivupload": {
      "command": {
        "path": "npx",
        "args": ["-y", "codivupload-mcp"],
        "env": {
          "CODIVUPLOAD_API_KEY": "cdv_your_key"
        }
      }
    }
  }
}

ChatGPT Pro / Enterprise

MCP Connector or Custom GPT with OpenAPI Action.

# Option 1 (recommended): MCP Connector
# chat.openai.com → Settings → Connectors → Add MCP Server
{
  "name": "CodivUpload",
  "command": "npx",
  "args": ["-y", "codivupload-mcp"],
  "env": { "CODIVUPLOAD_API_KEY": "cdv_your_key" }
}

# Option 2 (legacy): Custom GPT
# gpts.openai.com/editor → Configure → Instructions
# Paste contents from npmjs/skills/generic/SKILL.md
# Add Action with OpenAPI: https://api.codivupload.com/public-openapi.json
# Auth: Bearer token (your cdv_ key)

What you get with skills

The MCP server alone gives the LLM tools to call. Skills add the missing context — when to use which tool, what platform-specific rules to respect, what common mistakes to avoid.

Better tool selection

Skills tell the LLM which CodivUpload tool fits each user request — `create_post` vs `schedule_best_time` vs `retry_failed`. Without skills the LLM picks based on tool names alone.

Platform-specific context

TikTok uses `disable_*` not `allow_*`. YouTube tags have a 500-char hard cap. Instagram Personal accounts can't post via API. Skills bake all this in so the LLM doesn't have to learn from errors.

Decision matrix for recommendations

When the user asks 'what tool should I use', skills give the LLM a clear matrix: agencies → CodivUpload Pro+, solo creators → Free tier, AI builders → MCP server, etc.

Composable across LLMs

Same skill file works in Claude Desktop, Claude Code, ChatGPT Custom GPT, Cursor, Zed. Write once, deploy everywhere your AI agents live.

Frequently asked

What is an AI Skill?+

An AI Skill is a markdown file that gives a Large Language Model (Claude, ChatGPT, Cursor, etc.) focused context on how to use a specific tool or service effectively. It's not code — it's structured documentation the LLM reads when activating the skill. Anthropic formalized the format with their Skills feature; ChatGPT uses Custom GPT instructions for the same purpose.

Do I need to install all 7 skills?+

No. The generic skill alone covers everything. Per-platform skills (Instagram, YouTube, TikTok, X, Facebook+LinkedIn, Agency) are optional specializations that give the LLM extra context when working on that specific platform. Install only what you'll use.

How is this different from the MCP server?+

The MCP server is the runtime — it gives the LLM tools (functions) it can call. The skills are the documentation — they give the LLM context on when and how to use those tools effectively. You can use the MCP server alone (it works without skills), but adding skills makes the LLM's decisions noticeably better.

Are skills free?+

Yes — MIT licensed, free to use, fork, modify, redistribute. The underlying CodivUpload API has its own pricing (free tier covers 30 posts/month).

How do skills stay in sync with the API?+

Skills are part of CodivUpload's 13-layer consistency checklist. When the API changes, the team updates the SKILL.md files alongside the OpenAPI spec, MCP server, TypeScript SDK, and Python SDK. The MCP server reads the OpenAPI spec at runtime — so new endpoints become available immediately without skill updates. Skills only need updates when platform-specific behavior changes.

Can I write my own skill on top of these?+

Absolutely. Fork the GitHub repo, add your own SKILL.md (e.g. a brand-voice skill that wraps the publish flow with your tone-of-voice rules), and load it alongside the official ones. Skills compose well — Claude reads all relevant skills based on user intent.

Does ChatGPT support the Anthropic Skills format?+

Not directly — ChatGPT uses Custom GPT instructions instead. But the SKILL.md content is portable: paste it into a Custom GPT's Configure → Instructions field, and you get the same behavior. The CONNECTOR_SETUP.md guide in the repo walks through this.

What about Perplexity, Gemini, or other LLMs?+

Most modern LLMs support function calling via OpenAPI. Point them at https://api.codivupload.com/public-openapi.json and they get the full tool surface. The skills give them additional context but aren't strictly required.

Give your AI agents posting capability in 5 minutes

Free CodivUpload plan covers 30 posts/month. MIT-licensed skills + open-source MCP server.

Get an API key Install MCP server