Agent Skills

SocialKit Agent Skills are a drop-in instruction pack for coding agents. They teach an agent, like Claude Code, Cursor, or Codex, how to use every SocialKit endpoint: the parameters, the response shapes, and the common workflows.

Install them once and the agent stops needing the API explained in every prompt. It just calls the right endpoint and gets structured social data back.

The skills live at github.com/SocialKit-API/skills.

Skills or MCP?

Both surfaces sit on the same SocialKit REST API and use the same access key.

• Agent Skills fit coding agents that read skill files (Claude Code, Cursor, Codex). They call api.socialkit.dev directly. No MCP server is required.
• MCP Server fits tool-calling clients (Claude Desktop, MCP Inspector, custom agents) that discover typed tools over Streamable HTTP.

If your agent supports skills, start here. If it speaks MCP, use the MCP server. You can run both.

Choose your integration

Skills call REST directly; they do not require an MCP server. Use MCP when you want your client to discover typed tools.

Install

In any coding agent that supports the skills CLI:

npx skills add SocialKit-API/skills --all

The --all flag installs the full set without an interactive picker. To clone instead:

git clone https://github.com/SocialKit-API/skills.git

Authentication

The skills read your access key from the environment. Set it in your agent’s environment or a project environment file that your agent loads:

SOCIALKIT_API_KEY=your_access_key_here

Get a key from your project’s Access Keys page in the dashboard. The same key works for the REST API and the MCP server. The skills attach it as the x-access-key header on every request. Treat it like a password, and never commit .env.

What’s included

Installing with --all gives you seven skills:

Each skill is a SKILL.md file the agent reads on demand, plus an agents/openai.yaml descriptor.

Example prompts

Install the skills in your project terminal and configure your access key before trying these examples.

Claude Code: research a creator

Research the latest 10 Instagram Reels from @nasa. Identify their strongest
topics and opening hooks, with view counts and source links.

OpenAI Codex: build a competitor dashboard

Build a competitor-monitoring dashboard using SocialKit. Compare recent videos
from three YouTube channels by topic and views, with links to each video.

Cursor: build a customer-feedback report

Add a customer-feedback report to this app using SocialKit. Group comments
from selected Instagram posts into complaints, questions, and praise,
with supporting quotes.

SocialKit returns the data; your agent writes the application code or performs the analysis. The AI agents guide has more task examples and a platform capability table. Search is available for YouTube, TikTok, and Instagram Reels. Comment retrieval is available for YouTube, TikTok, Instagram, and Facebook; a comment count in X or LinkedIn post metadata does not include the comments themselves.

How it works under the hood

Every skill maps onto the REST API documented in this reference. For example, video-transcripts calls POST /youtube/transcript, POST /tiktok/transcript, and the other transcript endpoints depending on the URL.

Because the skills use plain REST, there is nothing to run and nothing to keep in sync. Most successful calls cost 1 credit, while AI and video endpoints follow the same per-request or per-started-minute tiers as REST. Failed calls are free, and your normal rate limits apply.

Credits and rate limiting

Skills consume credits identically to the REST API and MCP. The monthly credit allowance on your plan acts as a spend cap. See Authentication for key management and the API Reference for the simple per-endpoint tiers.