The author expanded an open-source MCP directory for AI agents to automate B2B supplier inquiries, adding 330 Korean factory forms to reach over 1,400 total and soliciting suggestions for future scraping.
Like most people experimenting with Hermes agent, mine started out doing the usual responsible adult chores—sorting my inbox, scraping morning tech news, and summarizing feeds. It worked great, but at some point, it just felt like an over-caffeinated cron job trapped inside my terminal. Then I had this random late-night lightbulb moment: what if our agents could actually step outside localhost and find real suppliers in the physical world? I was tinkering with a hardware side project and needed specs and pricing for actual parts. Obviously, factories aren't going to roll out shiny "Agent APIs" anytime soon. But almost all of them have had a slightly dusty, working HTML contact form for the last fifteen years. Instead of waiting for every company to adopt universal agent protocols, I figured we needed a practical bridge right now. So I started hacking together an open-source MCP directory as a passion project. To be clear on how it actually runs: this is built for targeted B2B RFQs, not a spam cannon that blasts 1,000 sites at once: You describe what you need (e.g., "Find precision CNC shops for a small run of custom aluminum brackets"). The agent queries the MCP directory and pinpoints the top 2 to 5 matching suppliers based on materials, tolerances, and capabilities. It pulls the form endpoints and language metadata, then drafts proper technical inquiries matching local trade etiquette. It halts and displays the payloads so you can review the 2–5 drafts before anything gets submitted. I already had ~1,100 global manufacturing forms indexed. Over the weekend, I scraped and verified small-to-mid precision manufacturing sites across South Korea (mostly precision CNC shops, custom molds, robotics parts, and PCB assembly). The scraping adventure was equal parts fun and cursed: Scraped ~450 candidate sites across industrial clusters. Instantly ditched over 100 of them: broken 2000s-era guestbooks, bloated Wix setups that choke headless browsers, and aggressive WAFs. Ended up with 329 clean, verified static HTML endpoints. That officially bumps the directory up to 1,434 verified endpoints. The schema parser is dialed in, and since this is an open community project, I want to index whatever physical supply chains you guys are actually curious about: Regions: Germany (mechatronics/tooling), Taiwan (semiconductors/PCBs), Japan, or Vietnam? Verticals: Industrial sensors, sheet metal fabrication, raw polymers, or packaging? Drop whatever country or niche component you dread researching manually, and I'll throw it into the scraping queue for the next batch! (repo and directory links in the comments)
A curated thread listing 10 GitHub repositories for web scraping, including Firecrawl, Crawl4AI, Browser Use, and others, covering everything from simple scraping to stealth tools and LLM-ready data extraction.
A developer is building 402 MCP servers for AI agents and seeks community feedback on which paid microservices would be most valuable for making agents more useful.
Manus AI introduces a connector for Similarweb's market intelligence data via MCP, enabling automated competitive research and presentation generation from natural language prompts.
The author built TrendsMCP, a unified API and MCP server that aggregates trend data from over 40 sources to enable AI agents to easily access and analyze trending information without managing multiple scrapers.
A developer shares an open-source desktop AI agent client called KainClaw, built with Electron, supporting MCP, multiple AI providers, tools, and background tasks. The project is still early and seeks community feedback.