How is a skill different from an MCP server?

An MCP server is a running process the agent calls at runtime to invoke tools and fetch data. A skill is static knowledge the agent reads into context before any tool call — API documentation, example workflows, prompt templates, and configuration recipes. Many products ship both: an MCP server for live tool calls and a skill for the surrounding knowledge agents need to use the MCP correctly.

How do I install an agent skill?

The standard install is `npx skills add /skills -g --yes` (the `-g` flag installs globally; remove it for project-scoped install). For example, `npx skills add codex-data/skills` or `npx skills add helius-labs/core-ai`. The CLI fetches the skill manifest from the repo, validates it, and registers it with all detected agent clients.

Does Hive ship an agent skill?

Hive's roadmap includes a `hive-intel/skills` GitHub repo with skill manifests for crypto-data workflows — token analysis, wallet analytics, DeFi yield comparison, security checks, prediction-market queries. The skill installs alongside the Hive MCP server and gives agents the prompt patterns and tool-routing knowledge to use Hive correctly. Until it ships, the equivalent install path is `npx -y hive-intelligence@latest init --all --browser` which auto-registers the MCP across detected clients.

What is an agent skill? | Agent Skills Definition for AI Agents

Q: What is an agent skill?

An agent skill is an installable package that gives an AI agent specialized knowledge — API schemas, tool definitions, example prompts, and operational instructions — for working with a specific service. Skills install with one command (npx skills add ) and become available across the agent's context. The pattern was pioneered by Anthropic for Claude and adopted by the broader agent ecosystem.

An agent skill is a self-contained installable package that gives an AI agent specialized knowledge of a specific API, service, or workflow.

The agentskills.io specification defines the manifest format. A skill includes:

API schemas and tool definitions for the agent to invoke
Example prompts showing how to use the API
Operational instructions and edge-case guidance
Configuration templates for client integration
Optional: pre-trained context the agent loads before tool calls

Skills install with one command: npx skills add <repo> -g --yes. The CLI fetches the manifest, validates it, and registers it with all detected agent clients (Claude Code, Cursor, Codex CLI, Gemini CLI, and others that implement the spec).

Skills vs MCP servers

These are complementary, not alternative:

MCP server — a running process the agent calls at runtime to invoke tools and fetch live data. Stateful, networked, authenticated.
Skill — static knowledge the agent reads into context before any tool call. API docs, example workflows, prompt templates.

Most modern dev-tool products ship both: an MCP server for live tool execution, and a skill for the surrounding knowledge needed to use the MCP correctly. Codex, Helius, and Alchemy all publish public skills repositories.

Hive's roadmap includes a hive-intel/skills repo for crypto-data workflows.

Skills vs MCP servers

Related terms