medium-clone
PublicOpen-source Medium clone: multi-format editor, AI co-writer, digests and learning paths. MIT.
Harakumo1 commitUpdated 2 downloads
Loading the files…
Open-source Medium clone: multi-format editor, AI co-writer, digests and learning paths. MIT.
curl -L https://nucleus.harakumo.com/api/repos/16/archive -o medium-clone.zipAnyone can download a public repo. Secret files are never in the download for people who cannot read them.
A blogging platform for people who write to teach: a multi-format editor, an AI co-writer that runs on open-source models, a personalised reading feed, curated news and research digests, AI-built learning paths, and an MCP server so AI assistants can read and write your blog.
Next.js 16 (App Router) · React 19 · MongoDB (Mongoose) · Tailwind CSS 4 · any S3-compatible bucket for media · any OpenAI-compatible model endpoint (local Ollama by default).
Writing
src/lib/sanitize.js, src/lib/mermaid-sanitize.js).AI co-writer (src/lib/ai.js, src/lib/aiProvider.js)
gpt-oss:20b) or any OpenAI-compatible
endpoint you configure, so with the defaults your drafts never leave your machine. AI cover
images are the exception: their prompt is sent to Pollinations.Reading and discovery
src/lib/feed.js)./pulse): an RSS digest of ~20 educational sources (src/lib/newsSources.json),
filtered by the model to keep evergreen, teachable pieces, each with a summary and key
takeaways./learn): learning paths (series) the model assembles from existing published
posts. It selects and orders posts; it never writes the lessons.Profiles
/profile/<username>) with two templates and configurable tabs
(articles, about, projects, experience, social links, contact). The contact form lands in
the writer's inbox (Dashboard → Messages).Automation and admin
src/lib/internalCron.js) fetches news and research, generates
topic ideas, writes and fact-checks AI posts, and builds learning paths. Each pipeline can be
paused, configured and triggered from /admin/automation, with run logs and a global
pause switch.AI assistants (MCP)
/api/mcp is a Model Context Protocol endpoint with OAuth (dynamic
client registration, /.well-known/oauth-authorization-server). Add
<your site>/api/mcp as a custom connector in an MCP-capable assistant and approve the
connection; tools include listing and creating posts, product launches, Pulse news and
research papers.mcp-server/): a stdio MCP server with ~30 tools (posts, AI writing,
feed, Pulse, research, bookmarks, comments, search, profile) that talks to your instance
with a personal API key from Dashboard → API Keys. See MCP server.Accounts and billing
src/lib/stripe.js are informational: the Writer plan is a
profile badge and is not enforced on features today.Requirements: Node.js 20.9+, a MongoDB database, and (for the AI features) either Ollama running locally or another OpenAI-compatible endpoint.
npm installcp .env.example .env.local # then fill in MONGODB_URI and JWT_SECRET at leastollama pull gpt-oss:20b # optional: the default local modelnpm run dev # http://localhost:3000ADMIN_EMAILS before signing up with it: that account becomes
an admin and gets /admin.STORAGE_* variables for anything public.HARAKUMO_API_KEY, emails (welcome, password reset) are skipped with a warning.npm test runs the sanitiser tests (Vitest); npm run lint runs ESLint.To run a production build anywhere Node runs:
npm run buildnpm start(next start prints a notice about the standalone output; it still serves the app. The
Dockerfile runs the standalone server directly.)
A Dockerfile is included (standalone output, port 8080):
docker build --build-arg NEXT_PUBLIC_APP_URL=https://your-domain.com -t blogwrite .docker run --env-file .env.local -p 8080:8080 blogwriteEverything is configured through environment variables; .env.example lists them all with
comments.
| Variable | Needed for |
|---|---|
MONGODB_URI | Required. All data. |
JWT_SECRET | Required. Session signing (32+ characters). |
NEXT_PUBLIC_APP_URL | Links in emails, sitemap, OAuth metadata, MCP URL. Build-time. |
ADMIN_EMAILS | Comma-separated addresses that become admins at sign-up. Empty by default. |
CRON_SECRET | Scheduled jobs and admin "run now" triggers (16+ characters). |
ENABLE_INTERNAL_CRON | false turns the in-process scheduler off on this instance. |
OLLAMA_BASE_URL, OLLAMA_MODEL | Local model (defaults: http://localhost:11434/v1, gpt-oss:20b). |
AI_BASE_URL, AI_API_KEY, AI_MODEL | Any other OpenAI-compatible endpoint; overrides Ollama. |
POLLINATIONS_API_KEY | Optional; higher rate limits for AI cover images. |
STORAGE_BUCKET, STORAGE_ACCESS_KEY_ID, STORAGE_SECRET_ACCESS_KEY, STORAGE_ENDPOINT, STORAGE_PUBLIC_URL | Media in any S3-compatible bucket. |
HARAKUMO_API_KEY | Transactional email through Harakumo Mail. |
STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, STRIPE_PRICE_ID_WRITER | Optional paid Writer plan. |
When CRON_SECRET is set, the server runs these in-process (UTC):
| Job | When | What it does |
|---|---|---|
fetch-news | every 4 h | Pulls the RSS sources and keeps the educational pieces. |
fetch-research | every 6 h | Pulls recent arXiv papers and summarises them. |
generate-topics | 00:00 | Suggests new post topics. |
generate-posts | 06:00, 18:00 | Writes, fact-checks and publishes AI posts and product concepts. |
generate-series | Sun 03:00 | Builds learning paths from published posts. |
AI posts and learning paths are published under the first address in ADMIN_EMAILS, else the
first admin, else the first user in the database, so set ADMIN_EMAILS before the first run. Pause
any pipeline in /admin/automation, or set ENABLE_INTERNAL_CRON=false. The same jobs are
also plain endpoints (GET/POST /api/cron/<job> with Authorization: Bearer <CRON_SECRET>)
if you prefer an external scheduler.
mcp-server/ is a small Node package (stdio transport) for desktop AI assistants:
cd mcp-server && npm installBLOGWRITE_API_KEY=bw_... BLOGWRITE_URL=https://your-domain.com node index.jsCreate the key in Dashboard → API Keys (the page also prints a ready-made config block). In an
assistant's MCP config, run node /path/to/mcp-server/index.js with those two variables in
its env.
BlogWrite is open source (MIT) and is published on Nucleus,
Harakumo's code home. Its emails already go through Harakumo Mail (src/lib/email.js, one
HARAKUMO_API_KEY).
BlogWrite is a Node server (Next.js standalone output, MongoDB, an in-process scheduler), so on Harakumo it runs as a server app (Pro plan and up), not a static site.
Create a project at harakumo.com/dashboard → Projects → New project.
Create an API key (Settings → API keys) and add it to the project's environment as
HARAKUMO_API_KEY, so password-reset and welcome emails are sent.
Add the rest of the environment (Project → Environment), from .env.example:
MONGODB_URI: a MongoDB database (include the database name in the path). MongoDB is
required for now; BlogWrite stores all its data there.JWT_SECRET: a long random string (openssl rand -hex 32).NEXT_PUBLIC_APP_URL: your project's public URL, and ADMIN_EMAILS with your address.STORAGE_* values. In production, images belong in a
bucket, not inline in the database.AI_BASE_URL (plus AI_API_KEY / AI_MODEL) to an
OpenAI-compatible endpoint, or leave AI off and set ENABLE_INTERNAL_CRON=false.CRON_SECRET for the scheduled jobs, and the Stripe variables for paid plans.Deploy from a folder with the CLI, or ask your AI assistant with the Harakumo connector:
npm i -g @harakumo/cliharakumo login --key <your API key>harakumo deploy <project id> .Every deploy is saved as a commit in the project's private Nucleus repo.
Notes:
ENABLE_INTERNAL_CRON=false on all but one.<your site>/api/billing/webhook.src/ app/ pages (public site, /dashboard, /admin) and API routes (/api/*) components/ editor (TipTap, code editor, Copilot chat) and shared UI lib/ auth, db, ai, feed ranking, storage/upload, email, sanitisers, scheduler models/ Mongoose models proxy.js route protection (Next.js 16 proxy) instrumentation.js starts the in-process schedulermcp-server/ stdio MCP server for desktop AI assistantsMIT © 2026 Charan Raj Kandula