Use Cases

MCP Markdown Sharing

mkdshare.DEV exposes a cloud-hosted MCP server so AI assistants and developer tools can publish, update, and annotate markdown documents without leaving the conversation.

What MCP is

The Model Context Protocol (MCP) is an open standard that lets AI assistants call external tools. When an MCP server is connected to your AI client, the AI can invoke its tools the same way it uses built-in capabilities — without you switching apps or writing API calls.

Why document sharing matters in MCP workflows

AI assistants produce long, structured markdown outputs — implementation plans, project briefs, research summaries, code reviews, specs. The problem is that these outputs live inside the chat window. Sharing them requires copy-pasting into another tool, which loses formatting and breaks the workflow.

With the mkdshare MCP server connected, your AI assistant can publish a document and return a shareable URL in a single tool call. The URL renders cleanly in any browser, supports inline comments from reviewers, and stays linked to the original document so it can be updated later.

Available MCP tools

create_document Publish a new markdown document. Returns a shareable URL. Supports visibility, allowed_emails, allowed_domain, expires_at.

get_document Read the full content of any document the caller has access to.

update_document Edit the title, content, or access settings of a document the caller owns.

list_documents List all documents owned by the authenticated user.

create_annotation Add an inline comment anchored to a specific passage in a document.

list_annotations Read all inline comments on a document.

request_device_auth Start a Google sign-in flow from inside a headless environment. Returns a short URL and code.

poll_device_auth Complete a sign-in started by request_device_auth and return an API token.

Authentication

Most tools require a Bearer token. Pass it in the Authorization header when connecting the MCP server.

Two tools — request_device_auth and poll_device_auth — require no token. They let an AI assistant walk a user through Google sign-in without leaving the conversation.

Supported clients

Claude Desktop Add the URL and token to claude_desktop_config.json.

Claude Code Run: claude mcp add --transport http --scope user mkdshare https://mkdshare.dev/mcp --header "Authorization: Bearer <token>"

Cursor Add in Settings → Cursor Settings → MCP.

Codex Add to codex.yaml under mcpServers.

ChatGPT Connect in ChatGPT settings.

Cowork Add to Cowork configuration.

Any HTTP MCP client Point to https://mkdshare.dev/mcp with the Authorization header.

Example workflows

Spec → shareable link

Write a spec in the conversation, then: "Publish this spec with mkdshare, restrict it to @company.com, and give me the link."

AI output → client review

After generating a deliverable: "Publish this as a mkdshare document and give me a link I can send to the client."

Collaborative annotation

After sharing: "Read the annotations on mkdshare document abc123 and summarize the requested changes."

Iterative update

After review: "Update the mkdshare document with this revised content, keeping the same URL."

Connect your MCP client

Get a token and follow the setup guide for your client.

Setup guide →