blog · April 3, 2026
MCP Tools vs Agent Skills
Two different ways to extend what an AI agent can do.
MCP tools give agents new capabilities — functions they can call, APIs they can access, actions they can perform. They extend what an agent can do.
Agent skills give agents new knowledge — instructions, workflows, and domain expertise. They extend how an agent approaches a task.
The analogy
Think of it like a surgeon:
MCP tools = the scalpel, the monitor, the instruments
Agent skills = the surgical training, the protocol, the expertise
You need both. Tools without skills get the wrong thing done efficiently. Skills without tools can't act on anything.
Side by side
| MCP Tools | Agent Skills | |
|---|---|---|
| What they add | Capabilities | Knowledge |
| Form | Functions / APIs | Markdown instructions |
| Called by | The agent (tool call) | Loaded into agent context |
| Example | "Search the web" | "How to write a cold email" |
| Where defined | MCP server code | SKILL.md file |
| Invoked registry? | No | Yes |
How Invoked fits
The Invoked registry is an MCP server — but it delivers agent skills, not tools. When your agent connects to Invoked via MCP, it gains access to a searchable library of SKILL.md files.
The agent uses MCP to query the registry; what it gets back is a skill (instructional content), not a callable function.
# further reading