By Max Techera · Open-source (MIT) · Updated June 2026
Blog

Iterating with agents: the loop behind the output

Most agent value is in the iteration loop — generate, view, refine. Sharing the artifact each pass matters.

npxnpx drops-cli report.html --managed

Agents work in loops: emit HTML, you look, you steer, repeat. A publish-and-replace primitive (same URL, new content) keeps a stable link across iterations so reviewers always see the latest.

Replace-in-place

update_site overwrites a drop in place — the link never changes while the content evolves. Ideal for agent loops.

FAQ

Does it work with any AI agent?
Yes. It's a CLI plus a SKILL.md and an MCP server, so Claude Code, Codex, Cursor, OpenCode, Gemini CLI — anything that runs a shell or speaks MCP — can use it.

Share what your AI builds — on your own domain.

Open-source, zero-knowledge, free. Try it in one command:

npx drops-cli report.html --managed