Skip to content

A Markdown viewer for Claude Code, Codex and Cursor on Windows

By Kunal Sahu · Last updated

Short answer: ask your agent to write its plan or spec to a .md file, open that file in a Markdown viewer with live reload, and snap it beside your terminal. Every time the agent saves, the document re-renders with diagrams and highlighted code, and you keep your place.

The Markdown files coding agents write

  • Claude Code reads project instructions from CLAUDE.md and will write plans, design notes and docs to any .md file you ask for (Anthropic docs).
  • OpenAI Codex uses AGENTS.md for repository instructions and writes plans, summaries and READMEs as Markdown.
  • Cursor keeps project rules as Markdown-based files and its agent writes plans and docs straight into your repo.

These files are long, and they're dense with headings, checklists, tables, code and Mermaid diagrams. In a terminal or Notepad they're a wall of symbols.

A workflow that works

  1. Give the output a home. Ask for the result in a file, for example: "Write the plan to docs/plan.md before changing any code."
  2. Open it once. Double-click docs/plan.md (or drag it onto MD Reader). You don't need to reopen it again.
  3. Snap it beside the terminal. Press Win+← on the terminal and Win+→ on the viewer.
  4. Read while it writes. Each save re-renders in place. Your scroll position stays put, so you can keep reading section two while the agent appends section five.
  5. Reply with specifics. Because headings and checklists are readable, it's easy to answer "change step 3 and drop the Redis option" before any code is written.

Prompts that produce readable docs

Write the plan to docs/plan.md with:
- a one-paragraph summary at the top
- a Mermaid flowchart of the request flow (quote all labels)
- a task checklist using - [ ] items
- a "Risks" section with at most five bullets
Don't edit code until I approve the plan.

Asking for a summary first and a checklist last gives you an answer-first document you can approve in a minute.

Keep agent docs organised

After a few weeks, plans pile up. A simple convention keeps them findable and easy to diff:

docs/
  plans/
    2026-10-18-auth-refactor.md
    2026-10-21-billing-webhooks.md
  decisions/
    0001-use-sqlite.md
  README.md

Tell the agent the convention once, in CLAUDE.md or AGENTS.md: "Write new plans to docs/plans/YYYY-MM-DD-topic.md." Commit plans with the code they describe, so the reasoning travels with the change.

A five-point check before you approve a plan

  1. Does the summary match what you actually asked for?
  2. Does the diagram show every system the change touches?
  3. Are risky steps (migrations, deletes, auth) called out on their own?
  4. Is there a test or verification step for each change?
  5. Is anything in the checklist vague, like "clean up" or "improve"? Ask for specifics.

Why a dedicated viewer instead of your editor's preview?

Editor previews work, but they share space with the code you're also reviewing and need an extension for Mermaid. A separate reading window gives the plan its own screen. It's light enough to leave open all day next to the IDE, terminal and agents, and it works offline, so private specs never leave your machine.

Common questions

Does live reload work when the agent rewrites the whole file?

Yes. MD Reader watches the file on disk, so it doesn't matter whether the agent appends or rewrites. It re-renders after each save.

Can I open CLAUDE.md or AGENTS.md?

They're ordinary Markdown files, so yes. It's a quick way to check what instructions your agent is actually following.

More: fixing Mermaid diagrams that won't render and making .md files open by default.