PasteAI: One MCP Tool Call, One Shareable Link

Have you ever asked Claude to write up an analysis, then had no good place to put it?

I asked Claude to pull together a summary from a long Slack thread on a production incident. It wrote up the timeline, root cause, and action items in about thirty seconds. But it was sitting in a chat window, and sharing it meant copying the markdown somewhere else and hoping the formatting survived.

That is what prompted me to build PasteAI.

The Problem

When Claude produces something useful, the output lives in a chat window. If I want to share it, I am choosing between losing the formatting in a paste box, making the reader sign in somewhere, or screenshotting it and throwing away the text. None of those are right for a document that took thirty seconds to generate and five minutes to find a home for.

What I actually want is simple: Claude publishes a document, I get a link, and the link just works. No login for the reader, no third-party service, no reformatting.

What PasteAI Does

Think of it as a pastebin, but with properly rendered documents. Claude calls a tool, and the result is a clean URL with syntax highlighting and a readable layout, not raw markdown in a paste box.

Claude has two tools: publish_document, which takes a title and markdown content and returns a URL; and list_documents, which returns recent public documents.

The URL opens a rendered document with a themed UI and dark mode. The theme is saved in localStorage.

Documents are stored as plain markdown files in ~/.pasteai/documents/, with a metadata index at ~/.pasteai/documents.db. The content stays as plain files so you can grep them, back them up, or open them in any editor without the server running.

PasteAI is built for local use first, which makes it practical for solo developers and small teams without any infrastructure overhead.

Scripts and external tools can push documents without Claude via the REST API; protect writes with an API key.

Getting Started

The quickest way to install is:

curl -sSL https://raw.githubusercontent.com/pasteai/pasteai/main/install.sh | sh

This sets up the binary and configures Claude Code automatically. PasteAI is also available via Homebrew, Scoop, and on the releases page as .deb, .rpm, and Windows packages.

Run pasteai setup if you need to reconfigure, or pasteai doctor to check that everything is connected.

Three Modes

PasteAI can run in three modes depending on how you want to use it.

By default, when Claude Code launches the MCP server, PasteAI also starts an HTTP server inside the same process. No separate setup needed. This is embedded mode, and it is what I use day to day. The server runs while Claude Code is open and stops when you close it.

For a persistent server, run pasteai serve directly or via Docker Compose and point the MCP client at it with PASTEAI_URL. For sharing across devices or teams, deploy with -base-url and use Tailscale for private access.

The code is at github.com/pasteai/pasteai, MIT licensed. PRs are welcome.

If you are spending time finding a home for AI output that should take seconds to share, this might help.

How are you currently handling Claude’s output when you need to share it, what does your workflow look like?

Also available on Medium