TL;DR: I have a small open-source repo (
mcp-directory, a directory of MCP servers for Claude). It sat at 0 stars for weeks despite weekly traffic. On 2026-05-26 I opened the first GitHub Discussion on it — not a "we shipped X" announcement, but an honest question. Within 8 hours the repo had its first ever star. n=1 isn't proof, but the pattern is dirt cheap to replicate, so I'm writing it down.
The repo
LuciferForge/mcp-directory — an attempt to make MCP server discovery less painful. 8,922 servers indexed as of writing. npm package, weekly downloads in the single digits.
For weeks: 0 stars. Decent traffic on the npm page. Zero engagement on the repo itself.
What I'd been doing wrong
Every README, every blog post, every dev.to article was a methodology dump. "Here's how I built it." "Here's the architecture." "Here's what I learned."
Nobody asked.
A repo with 0 stars dumping its methodology on a hypothetical audience reads exactly like what it is: someone shouting into a room with the lights off.
What I changed
Opened the first ever GitHub Discussion on the repo. Title:
"What would make MCP-server discovery actually useful for you?"
Not "here's the directory we built." Not "check out our methodology." Just: we have a thing, we don't know if it solves your problem, tell us what your problem actually is.
The body had three specific prompts:
- How are you currently finding MCP servers?
- What's the single most frustrating thing about that process?
- If you could query this directory in one specific way, what would it be?
That's it.
What happened
- 8 hours later: repo's first ever star landed.
- 24 hours later: still 1 star, but two organic visits from search referrers I'd never seen before.
- Lesson: I changed exactly one variable (open a Discussion that asks instead of announces) and the metric that hadn't moved in weeks moved. Not causal — n=1. But the experiment cost me 15 minutes.
Why I think this works (theory, not proof)
A README is a brochure. A Discussion is a doorway. People who would never star a brochure will star a doorway, because a doorway implies someone on the other side who might actually listen.
The "first Discussion" specifically matters because:
- It signals the repo is alive (most dead OSS repos have 0 discussions)
- It signals the maintainer is listening (most live OSS repos have discussions that are bug reports, not conversation)
- It's cheap social proof for the next visitor — "someone is engaging with this"
The template (steal this)
If you have a small open-source project that's been quiet:
- Enable Discussions if they're not already on (
Settings → Features → Discussions) - Open thread #1 with a title that asks, not announces
- End it with 2–3 specific prompts so people know how to respond
- Reply to every comment within 24 hours, even one-liners
That's the whole template. Half my portfolio (polyscope, pnl-truthteller, polymarket-mcp) is going to get this treatment over the next two weeks. I'll publish the followup numbers regardless of whether they confirm or kill the theory.
What I'm not claiming
- That this scales. n=1.
- That stars matter. They don't, except as a leading indicator that someone cared enough to do a single click.
- That this works on repos people are actively hostile toward. (My repos are unknown, not hated.)
What I am claiming: if you're sitting on a 0-star repo and waiting for "marketing" to feel less gross, opening an honest question thread costs 15 minutes and is the lowest-shame distribution move I've found.
Live polyscope data (the sister project): 15,700+ Polymarket markets / 15.5M+ price snapshots / 60+ days at api.protodex.io/stats — same pattern, also getting the first-Discussion treatment this week.
If you want the dataset behind polyscope (free CSV + paid Pro snapshot at $5): gumroad.com/l/polymarket-pro
(Built solo, no agency, no funding. Comments + counter-experiments very welcome.)
Top comments (0)