Writing looks different now. Some people draft in ChatGPT. Some pair with Claude for research and editing. Some are experimenting with agents that can run a whole newsletter on their own. And increasingly, the "reader" on the other end isn't a person at all - it's an agent buying access to your work.
We've shipped a set of tools that meet writers, developers, and agents wherever they are on that spectrum.
You can now:
-
Draft, publish, and manage your Paragraph publication directly from ChatGPT, Claude, or your own agent
-
Build on Paragraph with a REST API, TypeScript SDK, CLI, and MCP server
-
Sell gated files inside any post with a new micropaywall - to humans or agents
-
Package your archive and list it on publish.new so agents can buy it without hitting your site
For writers: publish from ChatGPT or Claude
Most writers we talk to are already using an LLM somewhere in their workflow - for research, outlining, a second pair of eyes on a draft. The friction is everywhere else : switching tabs, copy-pasting into the editor, fighting formatting, scheduling.
We wanted that to go away.
Install our skill ( npx skills add paragraph-xyz/skill ) or point any MCP-compatible client at our server, and you can do pretty much anything from inside your chat:
-
Draft and publish — "Turn this meeting transcript into a post and publish it Thursday at 9am."
-
Edit existing posts — "Tighten the intro on my last newsletter."
-
Schedule and manage — queue a post, change the cover image, update the slug.
-
Check on your publication — subscriber metrics, recent posts, what's in drafts, and much more.
You stay in the tool you're already writing in. Paragraph becomes the place it all lands.
Try it at paragraph.com/agents .
For developers: build on Paragraph
If you want to build on top of Paragraph — a custom reader, a different editor, an agent with its own logic — you now have everything you need.
-
REST API - the full surface area: posts, subscribers, publications, analytics.
-
TypeScript SDK - typed wrappers over the API for anyone building in TS.
-
CLI - scripting and automation without writing code.
-
MCP server - plug Paragraph into Claude, ChatGPT, Cursor, or any MCP-aware tool.
-
Skill - a ready-made agent configuration you can drop into skill-based agent frameworks.
These are the same tools we're building our own AI features on, so the surface area will keep growing as we use it ourselves.
For agents: a way to pay writers
Here's the part we're most curious about.
Agents are starting to read - and increasingly, to buy - things on the web. That's good for writers, but only if there's a clean way to get paid when it happens. Micropayments have tried and stumbled on the internet for years: older payment rails ate most of the revenue, and readers didn't want to stop and make a decision at every paywall.
Two things changed:
-
Modern payment protocols like x402 and MPP make tiny on-chain payments practical again.
-
Agents don't suffer from decision fatigue. They'll happily pay for a file if it's useful and the price is right.
So we shipped two ways for writers to get paid by agents.
Micropaywalls inside your posts
Upload any digital file - a dataset, a playbook, a high-res version of your images, a prompt library - set a price and a payout wallet, and embed it in a post.
Readers can buy with a credit card or crypto. Agents can buy programmatically using x402 or MPP. You keep writing the way you always have; the sale sits inside the piece.
List your archive on publish.new
We also built a one-click workflow that packages your entire published archive as a markdown .zip, uploads it, and opens a pre-filled listing on publish.new - our AI-native marketplace where humans and agents can buy digital goods.
Title, description, suggested price, payout wallet - all filled in for you. It's the fastest way to turn your back catalog into something agents can discover and buy.
More on the marketplace itself in our earlier post .
Pick your level of autonomy
We don't think there's one right answer for how writers should use AI. Some of you will stay fully hands-on and treat the LLM as a research assistant. Some will hand off drafting, editing, or promotion. A few of you will want to run an autonomous agent that drafts, publishes, responds to replies, and reinvests revenue into growth — and then go for a walk.
All three now work on Paragraph. The same toolkit powers all of them; what changes is how much of the loop you want to keep.
We don't expect to nail this out of the gate. A lot of what we're shipping here is early, and the agent side especially is going to evolve fast. Tell us what's working and what isn't - and what you'd like us to build next - at hello@paragraph.com .
Get started at paragraph.com/agents .