Markloop

Coding agents like Claude Code and Codex now draft specs, reports, proposals and technical docs as HTML, but getting human feedback on that work is broken. Pasting an HTML artifact into Google Docs destroys layout, code blocks and structure. Feedback ends up scattered across Slack, email and screenshots, unanchored to the actual line or section it refers to. The person who ran the agent becomes a middleman, manually translating comments back into prompts. Claude Code artifacts are also stuck inside a claude.ai org, so people outside it cannot even view or comment on them.

Markloop is a web app where agent-made HTML files become versioned 'projects' that reviewers can open and comment on directly, pinned to an exact section or sentence, without needing an agent or account themselves. Creators share a project or a single read-only file link with access, privacy and expiry controls, and reviewers see only the rendered document, never the source file. Comments, quotes, targets and context export as an agent-ready package (a comments.md plus the HTML file), or sync natively over MCP for Claude Code and Codex, so an agent can pull open feedback, apply changes locally, and publish the next version. Versions track which comments are addressed versus still open, and creators can also embed direct questions for reviewers to answer in place.

Coding agents now routinely produce full documents like specs, RFCs and reports, not just code, and Claude Code artifacts are locked inside an org with no external commenting, creating a gap Markloop fills for sharing and reviewing that output.