I spent years watching founders — myself included — treat LinkedIn like a chore they’d get to eventually. The problem was never a lack of ideas. It was that every “AI LinkedIn tool” on the market wanted a subscription, an API key, a dashboard login, and access to your content calendar sitting on someone else’s server. So I built the version I actually wanted: something that runs entirely inside Claude Code, costs nothing beyond what you already pay for Claude, and never asks you to hand your content strategy to a third-party app.

It’s called linkedin-growth-ai-agent, and it’s fully open source: github.com/rajendra-celestial/linkedin-growth-ai-agent.

The problem with every other LinkedIn AI tool

Nearly every LinkedIn content tool I evaluated before building this one follows the same pattern: sign up, connect your LinkedIn account, pay monthly, and trust a hosted backend with your positioning, your drafts, and often your actual account credentials. For a founder already juggling product, sales, and hiring, adding another SaaS subscription and another login to manage is exactly the kind of overhead that gets a tool abandoned within a month — not because it doesn’t work, but because using it becomes its own chore.

The deeper issue: most of these tools are a thin prompt wrapped around a generic AI call, charging monthly for something that’s mostly just prompt engineering you can’t see or edit.

What “no app, no database” actually means

This isn’t a metaphor for “simple to use.” There is genuinely no server running anywhere. The entire system is a folder of markdown instruction files — what Claude Code calls skills — that Claude reads during your conversation and follows. When you ask it to write a post, Claude reads the `content-writing` skill’s instructions, generates the post following that structure, and writes the result to a plain markdown file in your own `/content` folder. Nothing leaves your machine except your conversation with Claude itself, which you already have anyway.

This matters more than it sounds. Every other approach to “AI content tools” duplicates infrastructure Claude Code already provides — a separate LLM API key, a database to store your drafts, a hosted app to maintain. All of that becomes unnecessary the moment you accept that Claude Code itself is the engine, and a well-organized set of instructions is the only other ingredient.

What’s actually inside — 22 skills, not one prompt

The repository ships 22 distinct Claude Code skills, each a focused instruction set for one specific job, so Claude automatically picks the right one based on what you ask for:

• Profile analysis — feed it your LinkedIn profile and it extracts your niche, ideal customer profile, authority level, tone, and content pillars, rather than asking you to define these abstractly yourself.
• Content writing — covers 39 distinct content types (daily posts, founder stories, carousels, polls, case studies, thought-leadership pieces, video/reel scripts), each generated with a hook, body, call to action, hashtags, and SEO keywords built in, not bolted on afterward.
• Content calendar — 30, 90, or 365-day themed rotation plans, so you’re not starting from a blank page every week.
• Hook and CTA libraries — curated, living collections in the content archive itself, meant to grow as you use the system, not a static list baked into the tool.
• GEO optimisation — a skill specifically for structuring LinkedIn content so it performs well when AI answer engines (the same category of tool reading this post) parse and cite it. This is the same principle behind writing this very article the way it’s written.
• LinkedIn SEO, competitor analysis, audience research, trend research, cold outreach, DM writing, newsletter writing — the full list is in the repo’s architecture doc, but the shape is consistent: each skill does one job well instead of one skill trying to do everything passably.

Why files instead of a database

The content archive — every generated post, your profile analysis, the hook and CTA libraries, your calendar — lives as plain markdown files in a `/content` folder. Not a database table you need a tool to query. This was a deliberate choice, not a limitation: markdown files are readable without any special software, diffable in git so you can see exactly what changed and when, and portable to literally anywhere else (Notion, a spreadsheet, a future app) without a migration step. A database gives you none of that unless you build the export tooling yourself — tooling every hosted LinkedIn AI tool would have to build and maintain, and none of them prioritize.

What using it actually looks like

There’s no onboarding flow, no settings page, no dashboard. You open the folder in Claude Code and talk to it the way you’d talk to a strategist who already has context on your business:

“Analyse my LinkedIn profile.”

“Write me an educational post about hiring your first AI engineer.”

“Build me a 30-day content calendar.”

“Give me 5 comment ideas for this post I want to engage with.”

Claude matches the request to the right skill automatically and works from there. If you’ve used Claude Code for anything else — coding, research, writing — this will feel identical, because it is identical. The only difference is what instructions Claude is following.

What I deliberately left out

The roadmap doc is explicit about this: a hosted web app with its own AI provider integration and a Postgres database was considered and ruled out for this package specifically. Not because it’s impossible — because building it would mean re-solving problems Claude Code already solves, just to end up with a worse, more expensive version of the same output. If that kind of hosted product ever makes sense, it belongs as a separate thing built on top of this foundation, not baked into it from day one and carrying infrastructure most users would never need.

Why I’m giving this away

I built this tool for my own use first — I run content for Celestial IT Verse and for my own founder brand, and every existing option asked for more trust and more money than the problem justified. Once it worked well enough that I stopped thinking about it and just used it, open-sourcing it was the obvious next step. It costs me nothing to share, it costs you nothing beyond Claude Code access you likely already have, and it’s MIT licensed — fork it, strip it down, extend it, or just use it as-is.

If you’re a founder who knows LinkedIn matters but hasn’t found a system you’ll actually stick with, this is built to be the version that doesn’t ask you to manage one more login to find out.

Get it on GitHub →

Want help setting this up for your own brand, or figuring out what your content pillars should actually be before you start generating posts? I work with founders on exactly this — book a free 30-minute call and we’ll figure out what’s actually worth building for your situation, no pitch, no obligation.

Frequently asked questions