Every day, dozens of useful AI tools, MCP servers, frameworks, and services get published across GitHub, Reddit, HN, X, blogs, start ups and Product Hunt. Most developers miss 90% of them because nobody has
time to check every platform.
I kept running into this problem. My business partner who manually trawls these platforms would find tools that genuinely saved us hours on projects, things I never would have found on my own. So I built
a system to automate what he does.
What it does
https://claudecodetools.dev runs 5 autonomous AI agents on a VM that continuously scrape, classify, and score developer tools as they get published.
The 5 agents:
- Scout scrapes Reddit, GitHub, HN, X, blogs, and Product Hunt on intervals ranging from 15 to 120 minutes depending on the source
- Analyst classifies every discovery using Claude CLI, extracting the category, tags, pricing, install commands, and a novelty rating (well known, emerging, hidden gem, brand new)
- Enricher pulls GitHub metadata like stars, forks, last commit date, and whether the repo has tests
- Sentinel runs health checks every 6 hours to see if tools are still alive, declining, or dead
- Publisher generates expanded descriptions, pros/cons, and setup instructions for SEO pages
Every tool gets a legitimacy score from 0 to 100 based on GitHub signals, source credibility, engagement metrics, and cross referencing across multiple sources.
The toolkit generator
This is the feature I'm most proud of. You describe your project and it recommends non obvious tools you would never find by googling.
I tested it with "a Palantir style system that tracks space launches and objects in 3D" and it recommended:
- Three.js specific Claude Code skills for instanced rendering of 10,000+ satellites
- A semantic memory MCP server for persisting orbital data across sessions
- A governance hook that prevents Claude from rewriting your entire physics engine when asked to fix one method
- Token reduction middleware that stops Claude from re-reading your orbital math utilities every turn
Each recommendation comes with a "what you'd miss without this" explanation and a generated CLAUDE.md with actual install commands.
How I built it
The entire system was built in one session using Claude Code as project manager coordinating 3 parallel Claude Code sessions.
I wrote detailed instruction files (.md) for each session:
- Chat 1 (Backend): All scrapers, API routes, scoring algorithm, database schema
- Chat 2 (Frontend): Palantir inspired UI with sidebar filters, data tables, resource detail pages
- Chat 3 (Intelligence): RAG pipeline, classifier, toolkit generator, content generation
Each chat got one prompt: "Read CLAUDE.md and CHAT1_INSTRUCTIONS.md. Build everything listed there."
Then they ran simultaneously without conflicting because each had clearly defined file ownership.
After each round I would audit the results, identify integration gaps, write fixup instructions, and send them back. We went through about 6 rounds of this.
Architecture
Ubuntu VM ($30/month Azure)
-Next.js (App Router, port 3001)
-SQLite via Prisma
-Nginx (reverse proxy, SSL, security headers)
-PM2 (process manager, auto restart)
-Claude CLI (authenticated, used for classification)
-5 agents (auto start on boot via orchestrator)
The agents start automatically when the first API request hits the server. PM2 keeps everything running and restarts on crashes. Certbot handles SSL renewal.
Total infrastructure cost: about $30/month for the VM plus $12/year for the domain.
What's next
Currently at 330+ resources indexed across 14 categories and growing daily. The agents are finding about 10 to 20 new tools per day depending on what gets published.
Things I want to add:
- Community voting (thumbs up per resource, IP limited, no login needed)
- Weekly digest page showing what's new
- Security scanning of tools using a separate tool I built called Sentinel
- More scraping sources and broader coverage beyond Claude specific tools
Check it out at https://claudecodetools.dev and let me know what you think. Especially interested in hearing about tools you use that we might be missing.
Top comments (0)