Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseCode / SDK · ActionableX: @illyism→ original

Instead of Five Postiz Services, a Developer Wrote an MCP Server for AI Agents at $0 a Month

Ilyism wanted to hand post publishing to AI agents through the Postiz scheduler, but in one evening he wrote his own MCP server instead, a set of commands that Claude Code and Cursor can call. Instead of Postiz's five services (Next.js, NestJS, Temporal, Postgres, Redis), he runs a single Bun process with SQLite and no npm dependencies. Its 21 tools check platform limits, create drafts, and pull analytics. His conclusion: an agent needs a narrow API, and a calendar built for humans isn't necessary.

AI-processed from X: @illyism; edited by Hamidun News
Instead of Five Postiz Services, a Developer Wrote an MCP Server for AI Agents at $0 a Month
Source: X: @illyism. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceX: @illyism ↗
Editor-in-Chief
Key metric #126 TypeScript files, type checking in about 20 ms, access tokens encrypted with AES-256-GCM
Key metric #2Runs on a Hetzner server the author already pays for, at $0 a month, using a few megabytes of memory
Connects to 20+ platformsX, LinkedIn, TikTok, Telegram, and others

If you already have an AI coding agent such as Claude Code or Cursor, it's tempting to hand it the routine work around a product too: social media posts, post scheduling, and analytics. Usually people set up a ready-made scheduler like Postiz, an open-source service for scheduled publishing. Developer Ilyism took that route at first, but decided that running a whole stack just for an agent was too expensive.

Instead, in a single day he wrote his own MCP server. MCP (Model Context Protocol) is a protocol that gives an agent a set of ready-made commands: "check the limit," "create a draft," "schedule a post." Everything runs on Bun, a fast JavaScript and TypeScript runtime.

According to the author, the server needs no third-party packages or separate databases, and it costs nothing to run.

What Didn't Work for Postiz

Postiz is good for people: a calendar, previews, and team collaboration. But, according to the author's description, running it yourself requires five services: a Next.js frontend, a NestJS API, the Temporal task orchestrator, a Postgres database, and Redis. On top of that come thousands of npm packages and a heavy web calendar.

For a person, that setup makes sense. An agent doesn't need a drag-and-drop calendar. It needs clear operations: check whether text fits a platform's limit, upload an image, put a draft in the queue, schedule a publication, request statistics. These are exactly the operations that get packaged as MCP tools.

How the Replacement Works

According to the post, the whole service is a single Bun process with the built-in bun:sqlite database in WAL mode. Data lives in one file, and reads and writes don't block each other. No separate database, message broker, or orchestrator is needed. There are zero npm dependencies at runtime. The code is 26 TypeScript files, and type checking takes about 20 milliseconds.

Access tokens for social networks are stored in an encrypted vault using AES-256-GCM. Media files are temporarily placed in Cloudflare R2, from which the platforms pull them when publishing. The MCP server itself exposes 21 tools: the agent checks limits, prepares drafts, schedules posts, and pulls analytics.

The list of platforms is long: X, LinkedIn (including company pages and native PDF carousels), Instagram, TikTok with Photo Mode, YouTube Shorts, Threads, Facebook Pages, Pinterest, Reddit, Bluesky, Mastodon, Farcaster, Nostr, Lemmy, Dev, WordPress, Medium, Hashnode, Google Business, Slack, Discord, and Telegram. Later the author added press release feeds for SEO and for citations in ChatGPT and Google AI answers, encrypted direct messages on X, and carousels of up to 20 images.

The service runs on a Hetzner server the author already pays for and uses a few megabytes of memory, so the author estimates the additional cost at $0 a month. There are also no hundreds of third-party packages whose updates and vulnerabilities would need tracking.

What to Take Away and Where the Catch Is

First: tools for agents are better built around a narrow set of operations rather than around an interface designed for humans. The fewer layers there are, the easier it is for an agent to handle errors, and the fewer tokens it spends understanding its environment.

Second: the combination of Bun, SQLite, and a single process is a realistic choice for personal and small team automations that don't need to scale to thousands of users.

Now the limitations. This is the author's post on X, not an independent review. The post contains no link to open code, so the memory and speed figures can't be verified. A service written in an evening by one person may not have gone through everything a mature product faces: changes to platform APIs, rate limits, account blocks, multi-user permissions, fault tolerance. If you need a team with roles and a visual calendar, Postiz remains the logical choice.

The example is still instructive: when the main user of software becomes an agent, a good deal of heavy infrastructure turns out to be unnecessary. The author is already asking the community for the next step, an "alternative to Postiz," so an open version may yet appear.

Story

Key Questions Answered

What Didn't Work for Postiz?

Postiz is good for people: a calendar, previews, and team collaboration. But, according to the author's description, running it yourself requires five services: a Next.js frontend, a NestJS API, the Temporal task orchestrator, a Postgres database, and Redis. On top of that come thousands of npm packages and a heavy web calendar. For a person, that setup makes sense. An agent doesn't need a drag-and-drop calendar.

How the Replacement Works?

According to the post, the whole service is a single Bun process with the built-in bun:sqlite database in WAL mode. Data lives in one file, and reads and writes don't block each other. No separate database, message broker, or orchestrator is needed. There are zero npm dependencies at runtime. The code is 26 TypeScript files, and type checking takes about 20 milliseconds. Access tokens for social networks are stored in an encrypted vault using AES-256-GCM.

What to Take Away and Where the Catch Is?

First: tools for agents are better built around a narrow set of operations rather than around an interface designed for humans. The fewer layers there are, the easier it is for an agent to handle errors, and the fewer tokens it spends understanding its environment. Second: the combination of Bun, SQLite, and a single process is a realistic choice for personal and small team automations that don't need to scale to thousands of users. Now the limitations. This is the author's post on X, not an independent review.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…