Paperboy

0 stars
AGPL-3.0
Active this week
Last commit today

Last verified Nov 3, 2026

Generate fact-checked SEO articles by reading Search Console for topics, researching live sources, and holding any draft that fails the source comparison.

Visit website GitHub0 Deploy on Hostinger

Read the full breakdownWho it is for, how it works, trade-offs, the quick start and the FAQ

who it's for

Who Paperboy is for#

Content teams publishing weekly SEO articles

Paperboy reads Search Console, picks the topic with the best ranking opportunity, writes the draft, checks every claim against its source, and posts to WordPress or Ghost directly. A weekly Search Console report feeds back which topics performed, sharpening each subsequent topic pick.

Skip if:

Your articles are primarily opinion, news, or creative writing rather than fact-based SEO content. The fact check is designed for claim-heavy articles; it adds no value for pieces with no cited statistics or external references.

Developers integrating an article engine into an agent workflow

The MCP server and REST API expose the full pipeline as callable tools. Claude Code and Cursor can call suggest_topics, write_article, check_claims, and publish_article directly. write_article defaults to publish: false, so the agent asks before anything goes live.

Skip if:

You need a generate-and-publish pipeline with no hold step. Paperboy waits for approval by default; set PAPERBOY_APPROVAL_MODE to autopilot for unconditional publication on pass, but held articles still require manual review.

Solo operators who self-host and want no ongoing subscription

Self-hosting on Docker Compose is free. You pay only your Anthropic API bill, capped at $1.50 and 600,000 tokens per article by default, plus whatever a small VPS costs. There is no Paperboy licensing fee for the self-hosted path.

Skip if:

You want a managed service with one-click Search Console setup, a web dashboard, and a hosted blog on your subdomain. Those features are in Paperboy Cloud, not the self-hosted repo.

the problem

The problem it solves#

AI article generators write fast and publish confidently, but they cannot verify the claims they make. The source cited for a statistic may say something different, the competitor named may have wrong pricing, and the links may point to pages the model never read. A published error in an SEO article is hard to walk back; it spreads through search snippets, gets cached, and costs credibility that took months to build.

The harder part is that most tools have no review gate before publication. There is no record of which sources were read, no comparison of the article's numbers against those sources, and no hold for a failing draft. The review burden falls entirely on the writer, who typically checks a handful of sentences rather than every cited claim.

how Paperboy solves it

How it solves it#

Search Console topic selection

Reads your Google Search Console data to find queries where your site appears on page two, or gaps where no page on your site answers the query yet. Each suggested topic comes with its current position, impression count, and the reason it was chosen. A SERP-strength gate filters out topics where strong competing domains make ranking unlikely.

Source-save research pipeline

While researching, Paperboy reads up to six web pages and stores up to 24,000 characters of text per page. Every subsequent fact check runs against those stored texts, not the model's memory, so the check is repeatable and the saved source is inspectable after the fact.

Mechanical fact check before publication

Numbers in cited sentences must appear in the text of the cited page. Links must point to pages that were actually read during research. Sources that sell a related product are disclosed. Banned claims and house style rules run without a model, so they add no Anthropic cost.

Model judge for cited sentences

After rule checks, each cited sentence is scored by Claude Haiku 4.5 against the most relevant passages of its source: supported, partly supported, or not supported. A deterministic word-overlap fallback called RuleJudge runs with no Anthropic key. The judge also flags unsourced specific claims, opinions stated as facts, and voice drift.

Fail-closed publish gate

An article passes with zero blocking issues and a score of at least 70. A failing draft gets one automated fix pass on the flagged paragraphs, then the full check runs again. If it still fails, the article goes to needs_review status and is never published automatically. You can override via the API with force: true if you accept the risk.

MCP server for AI agents

Claude Code, Cursor, Codex, and any client that speaks MCP or HTTP can call the same engine the dashboard uses: suggest_topics, write_article, fact_check, publish_article, get_performance. write_article defaults to publish: false so the agent waits for your approval before anything goes live.

strengths · trade-offs

Strengths and trade-offs#

Strengths

Saved source text as the fact-check baselineMost AI writing tools rely on the model's training data or a live web search result. Paperboy saves the exact text of each source page at research time (up to 24,000 characters per page) and runs every number and citation check against those stored texts. The check is reproducible: you can inspect which source was saved and exactly where the discrepancy was flagged.
Fail-closed by defaultAn article that cannot pass the fact check goes to needs_review status and is never published automatically. You can override with force: true through the API, but the default is fail closed. Most paid AI writing platforms publish by default and leave error discovery to the writer.
Free self-hosting with per-article cost capsSelf-hosting on Docker Compose is free: you pay only your Anthropic bill. Per-article cost caps are set in the engine config at six web searches, $1.50 maximum, and 600,000 tokens maximum. No Paperboy subscription is required for the self-hosted path.
MCP and REST API for agent workflowsThe MCP server runs over stdio for self-hosted deployments or as a remote HTTP server for Paperboy Cloud. Any client that speaks MCP can call suggest_topics, write_article, check_claims, and publish_article without wrapping an HTTP client. Claude Code, Cursor, Codex, and Gemini CLI are documented with one-line install steps.

Trade-offs

-v0.1 alpha: early-stage projectThe project was created on 2026-10-29 and ships with an explicit v0.1 alpha badge. Production deployments should expect API changes between versions, gaps in edge-case documentation, and a small GitHub issue queue since the community is early. The self-host guide and contributing docs exist, but the codebase has not been tested at scale.
-Writing requires your own Anthropic key and Google OAuthWriting articles and the model judge both require an Anthropic API key. Running without one limits you to the deterministic RuleJudge and the paperboy check command only. Full Search Console integration also requires your own Google OAuth client, which adds setup steps beyond a plain Docker deploy.
-AGPL-3.0 restricts offering it as a hosted service to othersAGPL-3.0 allows self-hosting for your own sites and for paid client work on your own server at no cost. The restriction applies when you modify Paperboy and offer it to other people as a managed hosted service: in that case you must publish your changes under the same license. Running it yourself, including for clients, is unrestricted.

versus alternatives

Paperboy vs alternatives#

Paperboy vs Surfer SEO

Surfer SEO is a commercial content optimization platform that scores drafts against competing pages and suggests keyword adjustments. It does not write articles itself. Paperboy is an open source alternative that writes, fact-checks, and publishes articles directly from Search Console data.

Feature	Paperboy	Surfer SEO
License	AGPL-3.0	Proprietary
Self-hosting	Yes	No
Article writing	Yes (Claude)	No
Fact-checking	Yes, against saved sources	No
Search Console	Yes	No
Pricing	Free self-hosted	Subscription, no self-host

Paperboy suits teams that want an automated end-to-end pipeline: topic selection, writing, source verification, publishing, and Search Console feedback. Surfer SEO suits teams that already have a writing workflow and need SERP content scoring layered on top of it.

Paperboy vs Jasper

Jasper is a proprietary AI writing platform aimed at marketing teams. It writes fast across many formats, has no self-hosting option, and includes no fact-checking against saved source pages. Articles publish based on the model's output alone.

Feature	Paperboy	Jasper
License	AGPL-3.0	Proprietary
Self-hosting	Yes	No
Fact-checking	Yes, held if it fails	No
AI model	Claude (your key)	Proprietary
Search Console	Yes	No
Pricing	Free self-hosted	Subscription, no self-host

Paperboy is the better fit for teams publishing claim-heavy SEO articles where a verifiable source record matters. Jasper fits better when the use case is marketing copy, short-form content, or formats where per-claim source verification adds no value.

install · quick start

Quick start#

bash

Self-hosting deploys Postgres, the API server, and the worker with Docker Compose; edit the .env file with your credentials before starting.

```bash
git clone https://github.com/PaperboySEO/paperboy.git && cd paperboy
cp .env.example .env
docker compose up --build
```

tech stack · detected from GitHub

What it's built on#

Languages: JavaScriptTypeScript

frequently asked

FAQ#

Does Paperboy publish articles automatically or wait for approval?

By default, Paperboy waits. The PAPERBOY_APPROVAL_MODE environment variable defaults to approve, meaning an article that passes the fact check sits until you run paperboy publish . Set it to autopilot for automatic publishing of articles that pass the check.

What does the fact check actually verify?

Six checks run on every draft: every cited sentence has a source, numbers in cited sentences appear in the saved source text, the article says what its sources say (scored by a model judge), voice matches your site, no banned claims, and house style passes. Blocking issues each cost 12 points; warnings cost 2. An article needs zero blocking issues and a score of 70 or higher to pass. A failing draft gets one automated fix pass on the flagged paragraphs, then the full check runs again. A second failure holds the article in needs_review status until you review it manually.

Can I use Paperboy to fact-check articles I did not write with it?

Yes. paperboy check fact-checks any live page, and paperboy check draft.md checks a local markdown file. The exit code is 0 on pass and 1 on hold, so it works in CI. No Anthropic key is needed for the check command when the deterministic RuleJudge fallback is sufficient.

What publishing targets does the self-hosted version support?

WordPress and Ghost connect directly via their APIs. A signed webhook lets you push to any custom endpoint. A hosted blog on a subdomain and direct paste-in prompts are available on Paperboy Cloud but not in the self-hosted repository.

What does the AGPL-3.0 license mean for self-hosting?

Self-hosting Paperboy for your own sites, including internal team use and paid client work run on your own server, is free with no obligations. If you modify Paperboy and offer it to other people as a hosted service, the AGPL requires you to publish those modifications under the same license. Running it yourself is unrestricted.

also worth a look

Similar open-source tools#

Orkas

InAI & Machine Learning

Local-first desktop AI team: 16 agents, MIT, models of your choice.

6,249 GitHub stars
Licence MIT

Email Marketing Bible

InMarketing & Growth

Email automation skill for Claude, Codex, and ChatGPT agents.

Licence MIT

Claude Seo

InMarketing & Growth

SEO audit plugin for Claude Code, 26 skills and 19 agents

25,703 GitHub stars
Licence MIT

superCMO-skills

Alternative toAdCreative.ai

Generate video ads and product photos from inside your AI agent

270 GitHub stars
Licence Apache-2.0

Altorank

Alternative toSurfer SEO

Open-source SEO content engine with human-approval publishing

5 GitHub stars
Licence AGPL-3.0

newsjack

Alternative toMuck Rack

Open-source PR skills for AI agents. No agency required.

1,663 GitHub stars
Licence MIT

by category

Open source tools in these categories#

Marketing & Growth

agency-agents, MoneyPrinterTurbo, MiroFish 454

AI & Machine Learning

freeCodeCamp, REA (Reverse Engineer Anything), OpenClaw

Hostinger

Self-host Paperboy on a Hostinger VPS

Keep it running 24/7, even when your laptop is off. A Hostinger VPS gives it a server of its own.

01
Choose a plan
Starts from $6.49/mo
02
Apply
Extra 10% off yearly plans
03
Install Paperboy
On a server that stays on
04
Start building
Focus on the app, not the server

Deploy on Hostinger

Inbox Zero

Alternative toSuperhuman

AI email assistant that drafts replies and clears your inbox

12,889 GitHub stars
Licence AGPL-3.0
#Email

rakazo

Alternative toGrok Bot

AI teammates you own: your keys, your model, your machine.

6,568 GitHub stars
Licence Apache-2.0
#AI Agents

SnitchFeed

Live monitoring on LinkedIn, X, Reddit & more. Agentic social listening for startups.

Marketing & Growth

Advertise your product

The projects we add and every new guide. Free, unsubscribe any time.

Find open source alternatives to paid software

Paperboy vs Surfer SEO

Feature	Paperboy	Surfer SEO
License	AGPL-3.0	Proprietary
Self-hosting	Yes	No
Article writing	Yes (Claude)	No
Fact-checking	Yes, against saved sources	No
Search Console	Yes	No
Pricing	Free self-hosted	Subscription, no self-host

Paperboy vs Jasper

Feature	Paperboy	Jasper
License	AGPL-3.0	Proprietary
Self-hosting	Yes	No
Fact-checking	Yes, held if it fails	No
AI model	Claude (your key)	Proprietary
Search Console	Yes	No
Pricing	Free self-hosted	Subscription, no self-host