I built Sandesh — a content-publishing autopilot — solo, end to end. You draft a post once, and it ships to LinkedIn, Threads, Instagram and YouTube in your own voice, scheduled to land when your audience is actually awake. The whole thing is driven from Claude Code in the terminal.
The video above is a short walkthrough. Here's what's under the hood, and the decisions that made it worth building this way.
One MCP server per platform
Instead of a monolith wired to four social APIs, Sandesh exposes one MCP (Model Context Protocol) server per platform — LinkedIn, Threads, Instagram, YouTube. Each is OAuth-protected and hard-isolated, so a bad token or a rate-limit on one platform never takes the others down. Claude Code attaches to them as tools and orchestrates the entire publish flow.
Draft once, post everywhere
A single brief becomes per-platform output. Sandesh adapts voice and format to each surface — a long-form take for LinkedIn, a tight hook for Threads, a carousel or a vertical video where motion earns the scroll. One asset is uploaded once and reused across every platform instead of re-encoding it four times.
Human-in-the-loop publishing
Nothing posts on its own. Every draft renders into a preview-before-confirm envelope — captions, media, and the suggested time — and waits. I review it in the browser and hit confirm; only then does it fan out to the platform APIs. It's an autopilot with a pilot still in the seat.
One MCP fabric, five products
The same MCP + Supabase + channel-wake fabric that runs Sandesh also powers four other apps I have shipped solo: Setu (a browser to Claude Code relay), Sankalp (job-application autopilot), Swayam (scheduled automation), and Sathi (a personal manager). Build the plumbing once, reuse it five times — that is the whole point.
What was actually hard
The flashy parts were quick. The hard parts were the boring ones: keeping a single uploaded asset portable across four very different platform APIs, surviving expiring OAuth tokens, and making the publish gate genuinely safe — so an agent can move fast without ever shipping something I did not approve.
I'm open to work
Sandesh is one of five dogfooded products in my portfolio — proof that I can take an idea from zero to shipped, solo and in public. I'm looking for full-stack / AI engineering work: Next.js, Postgres, TypeScript, MCP servers, and LLM integration.
Book a free 30-minute consult →
Or reach out directly: theamargupta.tech@gmail.com
