Tools, Skills, MCPs, and Connectors Explained
Tools let agents act. Skills teach agents procedures. MCP standardizes tool/context access. Connectors link agents to apps and APIs.
Tools, Skills, MCPs, and Connectors Explained
Agent ecosystem has many words. These four matter most.
Tool
Tool = action agent can call.
Examples:
- read file
- search web
- run tests
- create GitHub issue
- send Telegram message
- query database
Tool should have clear input, clear output, and clear permission boundary.
Skill
Skill = reusable procedure.
It tells agent how to do task well.
Examples:
- code review checklist
- deployment workflow
- debugging playbook
- content writing template
- VPS troubleshooting steps
Tool gives capability. Skill gives method.
MCP
MCP means Model Context Protocol.
It is a standard way for agents and apps to connect to external tools, resources, and prompts.
Mental model:
- MCP client = agent app
- MCP server = exposes tools/data
- tool = action
- resource = readable context
- prompt = reusable instruction template
MCP matters because each agent does not need custom integration for every service.
Connector
Connector = integration with external app or platform.
Examples:
- GitHub
- Slack
- Notion
- Google Drive
- Gmail
- Linear
- Jira
- AWS
Connector may be implemented as MCP server, direct API tool, webhook, OAuth app, or automation platform action.
Quick comparison
| Concept | What it is | Example |
|---|---|---|
| Tool | action | run_tests() |
| Skill | procedure | “debug before fixing” |
| MCP | protocol | filesystem MCP server |
| Connector | app integration | GitHub connector |
Practical rule
Build in this order:
- Define workflow.
- Add minimal tools.
- Write skill/playbook.
- Add connector only if needed.
- Add MCP when reuse/interoperability matters.
- Add approval gates for risky actions.
Small safe tool beats powerful unsafe connector.