Every time I set up a new MCP server for Claude Desktop, I end up in the same loop: copy some JSON from a README, tweak the paths, fat-finger a comma, wonder why nothing works, stare at the config for 10 minutes, find the typo.
I figured there had to be a better way. There wasn't. So I built one.
What even is MCP?
If you haven't touched it yet — MCP (Model Context Protocol) is how AI apps like Claude Desktop, Cursor, and Windsurf connect to external tools. Want Claude to read your files? Search the web? Query your database? You need an MCP server, and that means editing a JSON config file by hand.
The config looks something like this:
{
"mcpServers": {
"filesystem": {
"command": "npx",
"args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me/projects"],
"env": {}
}
}
}
Simple enough for one server. But once you're running GitHub + Slack + PostgreSQL + Brave Search + a couple more, the config turns into a wall of JSON that's begging for a missing comma.
What I built
MCP Config Generator — a visual tool where you pick servers from a catalog, fill in your settings, and get valid config JSON. No signup, runs in the browser.
Here's what it does:
40+ server templates — Filesystem, GitHub, GitLab, Slack, PostgreSQL, MongoDB, Brave Search, Notion, Docker, and a bunch more. Each one pre-fills the command, args, and env vars so you don't have to dig through READMEs.
6 AI clients — Claude Desktop, Cursor, Windsurf, Claude Code, Cline, Continue. Each has slightly different config formats, and the tool handles that for you.
Custom servers — If your server isn't in the catalog, add it manually with whatever command/args/env you need.
Share links — Generate a URL that recreates your exact config. Useful for team setups or \"here's how to configure X\" docs.
100% browser-based — Nothing gets uploaded. Your API keys stay on your machine.
The catalog is organized by category (Filesystem, Dev Tools, Databases, Communication, Cloud, AI, Search, Productivity) so you can actually find what you need.
Why not just use the existing tools?
I looked. There are a few out there:
- Some are barebones text editors with validation
- Others only support Claude Desktop
- Most don't have a server catalog — you still need to know the exact npm package name and args
The gap I saw was \"I don't want to read 15 different GitHub READMEs just to configure my setup.\" Pick from a list, fill in your API key, done.
Try it
webtoolz.dev/ai/mcp — free, no account needed.
If you find a missing server or something that's wrong, the whole thing is open source: github.com/webtoolz-org
What servers are you all running? I'm curious what setups people have landed on — I'm personally using Filesystem + GitHub + Brave Search + Sequential Thinking as my daily stack.
Top comments (1)
The pain you're describing is the tip of a bigger iceberg.
Right now we have .cursorrules, CLAUDE.md, .github/copilot-instructions.md, AGENTS.md, and at least 8 other config formats — all doing roughly the same thing: telling AI how to work with your code. And none of them are portable.
A visual generator for MCP configs is useful, but the deeper question is: why does every tool need its own config format? Your workflow knowledge — the conventions, the patterns, the "don't do X" rules — should be tool-agnostic. When you switch from Cursor to Claude Code to Copilot, you shouldn't have to re-explain everything.
AGENTS.md is emerging as the closest thing to a standard (25+ tools already support it, 60k+ repos use it). Not because it's technically superior, but because it's the simplest: just a markdown file at your project root, readable by any tool.
Does your generator support exporting to multiple formats? That might be the killer feature — write once, generate configs for all your tools.