What is the problem with sharing a plan from an AI coding assistant?

The plan lives in your working tree or in the chat scroll-back. The reviewer is in Slack, Linear, or a pull request thread. Pasting 200 lines of markdown into chat breaks the tables and the mermaid diagram; committing a PLAN.md for a document that is obsolete the moment the code lands is noise in the repo. A share link is the middle path: the document renders at a URL with a table of contents, the diagram becomes a real diagram, and the link can expire when the review is over.

What does a good implementation plan look like?

This is a complete, fictional plan for adding webhook retries to a small payments API. It has the parts a reviewer needs: the decision being asked for, the problem, the proposed behavior, the alternatives that were rejected, a sequence diagram, the steps, verification, risks, and rollout. Download it, open it in the editor, or use it as a template for your own.

// scheduler/deliver.ts (sketch)
export async function deliver(attempt: Attempt) {
  const res = await fetch(attempt.url, { method: 'POST', body: attempt.payload, signal: AbortSignal.timeout(10_000) });
  if (res.ok) return markDelivered(attempt.id, res.status);
  const n = attempt.attempt + 1;
  return n >= 5 ? markDead(attempt.id, res.status) : reschedule(attempt.id, n, backoff(n));
}

How do I publish the plan from Cursor or Claude Code?

Connect the Docs MD MCP server once (per-editor configs are in the AI IDE guide). For Claude Code that is one command:

claude mcp add --transport http md-share https://docs-md.com/api/mcp

Then, with the plan open, ask:

Publish implementation-plan.md with a 7-day expiry. Give me the reading link,
and keep the edit token out of the document itself.

The assistant calls share_markdown and reports back something like this (the id and token are made up):

✓ Markdown shared successfully!

https://docs-md.com/quiet-harbor-7k2pd

Raw: https://docs-md.com/raw/quiet-harbor-7k2pd
Expires: 9/13/2026

Edit token (keep it to update or delete this share later): 4b1e…

Post the first URL in the review thread. Ask the assistant to note the edit token in the conversation or a local file that is not committed — it is the only way to change the share later.

How do I update the plan after review without changing the link?

Edit the source, then ask: "Update the share quiet-harbor-7k2pd with the new implementation-plan.md." The assistant calls update_share with the edit token. The URL you already posted now renders version two, and the expiry clock does not reset. From a script the same operation is:

curl -X PATCH https://docs-md.com/api/share/quiet-harbor-7k2pd \
  -H "Content-Type: application/json" \
  -H "x-edit-token: $EDIT_TOKEN" \
  --data-binary @<(jq -Rs '{content: .}' implementation-plan.md)

Where does the review discussion go?

Not on the shared page — Docs MD renders documents, it does not host comments. Keep the discussion where your team already works: the PR that will implement the plan, the Linear or Jira issue, or the Slack thread the link was posted in. The share is the readable artifact those conversations point at, and updating it keeps every pointer current.

What are the limits?

• Up to 120,000 characters of markdown per share, which is roughly 60 pages of plan.
• Expiry options are 1 day, 7 days, 30 days (the default), or never.
• Shares are public to anyone with the link; there is no reader authentication.
• Mermaid diagrams, GitHub-flavored tables, task lists, and highlighted code all render.

Related walkthroughs: hand a task to another coding agent and share an architecture diagram with notes. The full REST surface is on the API page.

FAQ