skills.sh has become the de facto directory for agent tools — over 40 agents use it. The ai.com.mcp/skills-search MCP server lets your agent query that directory at runtime, right when it needs a capability it doesn't have.
10-Second Pitch
- Runtime Discovery: Your agent queries skills.sh on-demand instead of hardcoding tool lists.
- 39+ Integrations: Works with Claude Code, Cursor, Copilot, Codex, OpenCode, and 35+ others.
- HAPI-Powered: Uses the HAPI protocol for efficient, structured skill retrieval.
- 2.5M+ Installs: Backed by the most-installed agent skill in the ecosystem.
Setup Directions
- Install the MCP server:
npx @agentskills/mcp-server - Add to your agent config:
skills add agentskills/agentskills - Set your API key:
export SKILLS_API_KEY=your_key - Query at runtime from your agent prompt
Example Prompt
"I need to send a Slack message with a chart attachment. Search skills.sh for a Slack MCP tool that supports file uploads and return the installation command."
Pros/Cons
| Pros | Cons |
|---|---|
| Dynamic tool discovery at runtime | Requires internet access |
| Cross-agent portability | Registry quality varies |
| Installs in one command | API rate limits on free tier |
Verdict: The skills.sh registry is where the agent ecosystem is heading. Having your agent query it on-demand vs. hardcoding 50 tools is the difference between a rigid bot and a genuinely adaptive one.