Back to search

Agent API

A free, read-only JSON API so AI agents can use the skill catalogue directly. No key needed. Each visitor can make 30 requests a minute and 500 a day; over that you get a 429 with a Retry-After header.

Endpoints

GET /api/public/v1/skills

Keyword search over the catalogue, most-installed first.

  • q: keywords, up to 100 characters (optional)
  • agent: restrict to one agent's curated list (optional)
  • source: match part of the GitHub owner/repo (optional)
  • minInstalls: minimum install count (optional)
  • limit: 1 to 50, default 20
  • offset: 0 to 5000, default 0
/api/public/v1/skills?q=react&limit=5

GET /api/public/v1/skills/{id}

One skill by its catalogue id, with its install command.

  • id: the id returned by a search
/api/public/v1/skills

GET /api/public/v1/skills/agent/{agent}

The curated list shown on one agent page, paged.

  • agent: microsoft-copilot, superhuman-go, chatgpt, grok, perplexity, claude, github-copilot, cursor, wordpress
  • limit: 1 to 50, default 20
  • offset: 0 to 5000, default 0
/api/public/v1/skills/agent/claude

GET /api/public/skills

Earlier flat-payload search, kept working for existing callers.

  • q: keywords, up to 100 characters (optional)
  • limit: 1 to 50, default 20
/api/public/skills?q=react&limit=5

Response shape

Version 1 endpoints answer with { data, meta } and, when something goes wrong, { error: { code, message } }. Paging details live in meta: count, total, limit, offset and nextOffset. Each skill includes id, name, description, source (GitHub owner/repo), installs, installCommand and url.

Agent connection (MCP)

Compatible assistants can connect to https://skills.mikedemo.dev/mcp over Streamable HTTP. No sign-in, read-only, same per-visitor limits.

  • search_skills — keyword search with agent, source and install filters
  • get_skill — one skill's details and install command
  • list_agents — the agents that have a curated collection

Machine-readable resources