MCP, API, skill or CLI: how to connect an agent to its tools

There are four common ways to give an agent access to a tool. They are not competitors: they suit different agents, and AMXP exposes the same catalog and balance through each one.

Skill: a file the agent reads

A skill is a markdown file that teaches the agent what a service is, when to use it and which commands to run. The agent reads it once and follows it. It needs no install on your side, which is why the AMXP install is one line:

Read https://amxp.ai/SKILL.md and configure AMXP.

Best for: coding agents and assistants that can run shell commands, such as Claude Code or Cursor.

MCP: tools the client lists for the model

The Model Context Protocol lets a client such as Claude or ChatGPT connect to a server and show its tools to the model as callable functions. AMXP's MCP server exposes three tools, amxp_discover, amxp_inspect and amxp_execute, rather than thousands, so the model's context stays small and it searches the catalog only when it needs to.

Best for: chat clients and agent frameworks with MCP support.

CLI: commands in a terminal

The amxp CLI wraps the same three operations as terminal commands. It suits scripts, cron jobs and agents that work through a shell. It also prints costs and run status in a form both people and models can read.

Best for: automation, CI and shell-first agents.

HTTP API: everything else

Under all three sits a plain HTTP API with an API key: POST /v1/discover, GET /v1/tools/{id}, POST /v1/runs. If your agent runs in your own backend, call it directly.

Best for: products that embed agents and want full control.

Which one should you pick?

Your agent	Use
Runs shell commands	Skill + CLI
Lives in a chat client	MCP
Runs in your backend	HTTP API
Not sure	Start with the skill; it tells the agent how to pick

Whichever you choose, it's the same catalog, the same prices and the same balance.