customer-discovery-agent

English | 中文

Scan online communities for maker pain points. Cluster into themes. Get a weekly digest of what your future users are actually frustrated about.

Built by Alex Ji — solo founder shipping VibeXForge, Orallexa, and marketing-agent. Born from this thought:

I keep guessing what indie makers actually need. Reddit knows.

What it does

customer-discovery-agent scan \
  --subreddits SaaS IndieHackers SideProject Entrepreneur \
  --hours 168 \
  --out digest.md

For each enabled subreddit, the agent:

1. Pulls top + new posts in the last --hours window
2. Filters by pain-pattern keywords (default: 17 phrases like "i wish", "frustrated", "is there a tool", "struggle with")
3. Sends matches to Claude Haiku for theme clustering (3-7 clusters)
4. Renders a markdown digest with representative quotes + sample URLs

Without ANTHROPIC_API_KEY, falls back to deterministic keyword-based clustering — still useful, just less coherent.

Sample output

# Customer Discovery Digest

*Window: last 168h · Sources: reddit*
*Pain points scanned: 47 · Clusters: 5*

## Top themes

### Founders frustrated with Vercel build minutes

**8 posts · avg score 42.5**

> $130 in build minutes this month, half from failed type-check pushes

Sample posts:
- https://reddit.com/r/SaaS/comments/...
- https://reddit.com/r/IndieHackers/comments/...

Install

git clone https://github.com/alex-jb/customer-discovery-agent.git
cd customer-discovery-agent
pip install -e .

Set Reddit credentials (PRAW script app at https://reddit.com/prefs/apps):

export REDDIT_CLIENT_ID=...
export REDDIT_CLIENT_SECRET=...
export REDDIT_USERNAME=...
export REDDIT_PASSWORD=...
export REDDIT_USER_AGENT="customer-discovery-agent/0.1 by yourname"

Optional: ANTHROPIC_API_KEY for LLM clustering.

CLI

customer-discovery-agent scan
  --subreddits SaaS IndieHackers SideProject ...   (required)
  --hours 168                                       (lookback window)
  --keywords "i wish" "frustrated" ...             (override default list)
  --min-score 5                                     (skip posts below this)
  --limit 100                                       (max posts per subreddit per stream)
  --max-clusters 7
  --out digest.md                                   (default: stdout)

Roadmap

• v0.1 — Reddit scraper + keyword filter + Claude clustering + markdown digest
• v0.2 — IndieHackers RSS, X timeline, 即刻 search, 知乎 question scraper
• v0.3 — Email shipping (Resend / Mailgun), GitHub Actions weekly cron
• v0.4 — Embedding-based semantic clustering (replaces LLM cluster call for cost)
• v0.5 — Cross-week diffing — "this theme grew 3x this week"

Why this exists

Solo founders make two systematic mistakes about user research:

1. Building from gut: skipping discovery → product-without-buyers
2. Doing it manually: scrolling Reddit for hours → either burns weeks or stops happening

This agent runs as a Sunday night cron, drops a digest in your inbox Monday morning, and your guesses become literal quotes from your future users. ~$0.001 / scan in template mode, ~$0.05 / scan with LLM clustering on a busy week.

MCP server (Claude Desktop / Cursor / Zed)

Ask "scan IndieHackers for pain points this week" inline from your AI assistant.

pip install 'customer-discovery-agent[mcp]'

{
  "mcpServers": {
    "customer-discovery": {
      "command": "customer-discovery-mcp",
      "env": { "ANTHROPIC_API_KEY": "..." }
    }
  }
}

Tools: scan(subreddits, hours, …) · latest_digest(directory) · keyword_list()

License

MIT — use it, fork it, add IndieHackers and 即刻 to it.

🧩 Part of the Solo Founder OS stack

A growing collection of MIT-licensed agents that share solo-founder-os as their base — Source/MetricSample contracts, HITL queue, AnthropicClient, notifiers, scheduler. Each agent is independently useful; together they cover the full solo-founder workflow.

Each agent's own row is omitted from its README. Install whichever solve real problems for you.