Mcp

If you read my last post, you know I built searxng-web-fetch-mcp — a Crystal MCP server that gives AI assistants web search and content fetching. It worked. It was fast. I shipped it in a night. But after a few weeks of daily use, things started to bother me. The extraction was good, but not great. Some articles came back with navigation junk. Others had the main content buried under sidebar text. And the batch fetch, while fast, didn’t give me the control I wanted over what came back and how. ...

It was 7pm on a Wednesday. I was staring at my terminal, watching OpenCode try to answer a question about a library it had never seen before. The LLM was doing its best. But it was hallucinating API endpoints that didn’t exist. And I thought: “Why can’t my AI just… search the web?” The Problem I use OpenCode, Claude Code and some times Crush as my daily coding companion. It’s powerful. But it has a blind spot: The native web fetch can’t access claudflare protected sites. ...

The last post was about architecture decisions. This one is about execution. I spent this cycle turning ideas into something runnable and testable. Not polished. Not “AI magic.” Just real foundations. What shipped since the last post 1) REST API vertical slice navi-agent now has a working API backbone with health, task, and agent routes, including sync flow. This gave me a full path from request → service → persistence → response, which is where real design flaws start to show up. ...

Hello, world! First of all, I want to say that this is my first post in a long time, and I hope to keep writing more often. I have been working on a MCP (Model Context Protocol) for a while now, and I have been learning a lot about typescript AST manipulation. I want to share some of my learnings with you. What is a MCP (Model Context Protocol)? MCP is an open protocol that allows you to standardize the way that applications provide context to their models. ...