Sociona · 2025–2026

Sociona

closed source

AI
SaaS
MCP
Next.js

Sociona landing page: manage all your social media in one space — www.sociona.app

2025–2026

A social workspace for short-form video creators: plan ideas, develop them into scripts, design and schedule posts across X, Instagram, Threads, TikTok, YouTube, LinkedIn and Facebook, and learn from your own videos. An API and two MCP servers let agents run the same workflows. In public beta.

read how it was built ↓

about.txt

role: Founder & Full-Stack
for: Sociona
stack: AI, SaaS, MCP, Next.js
when: 2025–2026

view live

How Sociona was built01 / 10

versionv0.1Sep 2025

The first commit

A multi-tenant SaaS from day one: Next.js on the front, a Node and TypeScript backend, PostgreSQL with strict tenant isolation, and OAuth to X, Instagram and Threads. Planner and poster agents wrote and published the posts.

scroll ↓ or click a card

The brief

Short-form creators juggle a notes app for ideas, an editor, a design tool for carousels, a scheduler for each platform and a separate analytics page for each. AI assistants can draft a caption but cannot publish it or see what actually worked. Sociona is one workspace for the whole loop: plan, develop, design, publish and learn. The same API serves the dashboard and agents.

facts.txt

Problem: The creator workflow is scattered across tools
Audience: Short-form video creators and small teams, plus developers
Platforms: X, Instagram, Threads, TikTok, YouTube, LinkedIn, Facebook
Surfaces: Web, mobile app, REST API, two MCP servers
Status: Public beta, 7-day free trial
Timeline: Sep 2025 to now

How I approached it

01
Start from the idea, not the post
An idea goes on a board, gets developed into a script in the Creation Studio, then scheduled. Publishing is the end of the flow, not the start.
02
One API, two front doors
The dashboard and the agents call the same endpoints. Anything a person can do in the app, an agent can do through MCP, and the other way round.
03
Learn from what you already posted
Cortex measures a video with code (shots, pacing, beats) and lets models label the hooks and structure. The result is stored, so screens reuse it instead of spending tokens again.
04
Make agents safe to hand the keys to
Dry runs before publishing, platform and account limits, approval holds, idempotent publishing and scoped API keys. Power came after the guardrails.
05
Design where the model can see
The Studio server returns renders as images and ships its design playbook as instructions, so a session builds, looks and fixes in a loop.
06
Long work runs in the background
Carousels, video analysis and renders run as queued jobs that survive restarts, and finish with a signed webhook instead of a held-open request.

A look at the dashboard

What sets it apart

Schedulers such as Buffer and Hootsuite are built around a queue: you bring the copy, they time it. General chatbots write well but cannot post, and know nothing about your accounts. Analytics tools show you numbers without telling you what to do next.

Sociona is built around the creator's loop instead:

One workspace for every platform, instead of a different app for each.
Ideas, scripts, designs and posts live together, so a good idea does not die in a notes app.
It analyses your own videos and points each insight at an action for the next one, instead of a wall of charts.
Agents can operate the whole thing through MCP, with guardrails, instead of being stuck at a draft.

What it runs on

Next.js 16Web app and a separate Studio app
Expo / React NativeThe mobile app
VoltAgent on HonoThe agents service, with the planner and poster agents
Node / TypeScript backendOne API for the apps, developers and agents
PostgreSQL + PrismaWorkspaces, accounts, posts and keys, isolated per tenant
Redis + BullMQScheduling, retries and long-running jobs
Better Auth + StripeSign-in and billing
Cortex (Python)Video analysis: PySceneDetect, ffmpeg, Gemini and pgvector
Python media service + render workerVideo processing and vertical renders
Fabric.jsThe carousel editor canvas
RailwayHosting

What got in the way

the problem
Giving an AI assistant the power to post to real accounts is risky.
fix.md
Posts can be dry-run and validated first, held for approval by profile guardrails, and published idempotently with a client request id so a retry never doubles a post.
the problem
Video models sample about one frame a second, so they cannot see fast cuts, zooms or beats.
fix.md
Cortex keeps a rule: code measures, models label. Shot boundaries, pacing and beat times come from deterministic analysis, and the model only labels hooks and structure on top.
the problem
Platform APIs shift under you. Instagram retired its basic display API, and the OAuth setups for Instagram and Google did not match the products they were meant for.
fix.md
Moved to the Graph API and wrote the setup, test-user and mismatch fixes down as docs, so connecting an account is a checklist instead of a mystery.
the problem
Carousels and video analysis take minutes, and long requests die behind proxies.
fix.md
Responses carry keepalives for multi-minute calls, and long tools can run as queue-backed jobs that survive restarts and finish with an HMAC-signed webhook, retried with backoff.
the problem
A model designing a carousel blind produces broken layouts.
fix.md
The Studio server returns each render as an image in the tool result, plus layout checks and an independent critique, so the model can look and fix before it calls the design done.
the problem
Platform media links are temporary or private, so they cannot go on a public site.
fix.md
Rendered media goes into the library with absolute, durable links, and the API says which platform covers expire so a site knows to cache them.
the problem
One video posted to three platforms looked like three videos.
fix.md
Cross-posts are grouped into one entry, with a link, handle and numbers for each platform.

What came out of it

2,572commitsSep 2025 to Oct 2026
10apps and servicesweb, studio, mobile, backend, agents, Cortex, media, render, docs
~70tools on the Sociona MCP serverup from 7 in the first release
~28tools on the Studio MCP serverdesign, with images in the results
7platformsX, Instagram, Threads, TikTok, YouTube, LinkedIn, Facebook

Sociona is in public beta with a hosted API, a mobile app and two MCP servers. These numbers describe what was built. There are no user figures to share yet.

What I learned

Build one API and let the UI and the agents share it. Every feature I shipped for people was automatically available to agents.
Guardrails before power. Dry runs, approvals and scoped keys are what make it reasonable to hand an agent real accounts.
Measure with code, label with models. A model that samples a frame a second should not be trusted with timing.
Let the model see what it makes. Images in the tool result changed the quality of generated designs more than any prompt tweak.
Long work belongs in jobs and webhooks, not in a request that someone has to hold open.

Sociona: a workspace for creators who actually post

In public beta with a 7-day free trial. Plan, develop, design, schedule and learn in one place, with MCP servers so agents can run the same workflows.

Try Sociona

Public beta. The MCP server is open source.

sociona.app MCP server

up next →

Clevaly

2025–2026

closed source

EdTech · AI · Mobile

A school workspace that keeps classes, deadlines, notes and study sets in one place, with Cleva, an assistant that has already read your syllabus. Live in public beta on iPhone, web and Chrome.

say-hi.txt

Want something like Sociona?

Tell me what you're making and I'll tell you how I'd build it.

Get in touch all projects

Overview

Sociona

The first commit

Bantr becomes Sociona

An MCP server on npm

Learning from videos

The Sociona app

The MCP platform release

A second server for design

Stencil, extracted

One video, three posts

Powering this site