Agent First Interface — scaffold tools whose primary consumer is an AI agent, not a human.

From a single source of truth, afi-cli generates three interface surfaces, each shaped by a different agent-ergonomic principle:

  • CLI — with a learn affordance so an agent can introspect the tool and author its own usage skill (not just read --help).
  • MCP server — a deliberately minimal menu, tuned for low surface area over maximal API coverage.
  • HTTP site — markdown pages plus a sitemap, navigable by any agent with a fetch tool. (This page is the HTTP surface.)

Quickstart 

uv tool install afi-cli
afi explain afi            # top-level map
afi overview               # cross-surface rollup
afi learn                  # structured self-teaching prompt for an agent

uv tool install is the supported install path — not pip install.

  • AgentCulture — the OSS org, its agents-as-members model, and where afi-cli sits inside it.
  • Agent First — the paradigm: learnability on the CLI, minimalism on MCP, discoverability on HTTP. The why behind every design call in this repo.
  • Rubric — the five-bundle mechanical check that afi cli verify runs against any CLI. afi-cli itself is required to pass.

afi-cli — the Agent First Interface scaffolder for the Culture ecosystem. Source on GitHub.

This site uses Just the Docs, a documentation theme for Jekyll.