I built the missing piece of the MCP ecosystem

Diwakar — Mon, 13 Apr 2026 06:00:17 +0000

I Built the Missing Piece of the MCP Ecosystem
If you've been building with AI agents lately, you've probably heard of MCP — Model Context Protocol. It's the open standard that lets AI agents like Claude connect to external tools: databases, APIs, file systems, code repos, payment systems. The ecosystem is growing fast. Companies like Stripe, Notion, Vercel, Redis, and Linear have all shipped official MCP servers.
But there's a problem nobody is talking about.

The Problem — MCP Servers Are Impossible to Discover
Here's what installing an MCP server actually looks like today.
You want to connect Claude to your PostgreSQL database. You Google "postgres MCP server." You find a GitHub repo. You read the README. You copy an npx command. You open your claude_desktop_config.json file manually. You paste the command into the right JSON structure, hoping you don't break the syntax. You restart Claude Desktop. You pray it works.
Now repeat that for every single server you want to install.
There is no central registry. No search. No standard install command. Servers are scattered across dozens of GitHub repos with no consistent format, no discovery layer, and no tooling. The MCP ecosystem has the servers. It just doesn't have npm.
That's what I built.

What I Built — MCPHub, the npm for MCP
MCPHub is an open community registry and CLI for MCP servers. The idea is simple: every MCP server in one place, installable in one command.
bashnpx mcphub install postgres
npx mcphub install github
npx mcphub install stripe
That's it. The CLI writes directly to your claude_desktop_config.json. Restart your AI client and the server is live with all its tools available.
The registry launched with 22 production-grade servers including official integrations from Stripe, Notion, Vercel, Redis, Linear, Qdrant, and the full official MCP server collection from the modelcontextprotocol team — filesystem, GitHub, postgres, memory, slack, puppeteer, brave-search, sqlite, git, and fetch.
Everything is MIT licensed. No accounts. No lock-in. Free forever.

How It Works — CLI + JSON Registry
The architecture is intentionally simple. The registry is a single servers.json file on GitHub. Every entry looks like this:
json{
"name": "postgres",
"author": "modelcontextprotocol",
"description": "Connect to any PostgreSQL database.",
"category": "Databases",
"install": "npx @modelcontextprotocol/server-postgres postgresql://localhost/mydb",
"tools": [
{ "name": "query", "description": "Execute a SQL SELECT." },
{ "name": "list_tables", "description": "List all tables in schema." }
]
}
The CLI is a TypeScript Node.js tool that fetches and caches this registry, lets you search it, and writes the install config directly to your AI client's config file. It handles the full lifecycle — install, list, remove, update.
The website is a zero-dependency single HTML file with full search, category filtering, dark mode, and a one-click install page. Deploy it anywhere. No build step, no framework, no dependencies.
CI automatically validates every community submission against a JSON schema before merging. Every server entry has a real GitHub URL, verified install command, and documented tool list.

Demo — Three Commands That Change Everything
Install the CLI once:
bashnpm install -g mcphub
Search the registry:
bashmcphub search "vector search"

Returns: qdrant, memory, and more

Install any server:
bashmcphub install memory

Writes to claude_desktop_config.json automatically

Restart Claude Desktop — done

You can also check what's installed, get full server details, and remove servers:
bashmcphub list
mcphub info stripe
mcphub remove slack
The whole workflow that used to take 10 minutes of manual config editing now takes 10 seconds.

Try It
The full project is open source on GitHub:
github.com/Diwakarsrd/MCPHUB
The live registry browser and one-click install page:
diwakarsrd.github.io/MCPHUB/
Star the repo if you find it useful. It genuinely helps with visibility and helps more developers find it.

What's Next — Submit Your Server
MCPHub is community-owned. The registry grows through pull requests. If you've built an MCP server — or know of one that should be listed — opening a PR is all it takes.
The entry format is simple, validation runs automatically in CI, and PRs get reviewed and merged fast.
Here's what I want to add next based on demand:

More database servers (MySQL, MongoDB, DynamoDB)
More productivity tools (Jira, Asana, Calendar)
Local AI tools (Ollama, local embeddings)
A mcphub init command to scaffold new MCP servers

If you build with AI agents and want a server added, open an issue or drop a comment below. I read every one.
The MCP ecosystem is growing faster than any developer tool I've seen. MCPHub is the infrastructure layer it's been missing. Come build it with me.

DEV Community: Diwakar

I built the missing piece of the MCP ecosystem

Returns: qdrant, memory, and more

Writes to claude_desktop_config.json automatically

Restart Claude Desktop — done