Background: I use Cursor, Claude Desktop, VS Code, and occasionally Claude Code.
Every time I wanted to add a new MCP server, I had to edit 4 separate JSON configs
and paste the same credentials in 4 places. When a token rotated, I'd always miss one.
So I built 1mcp — a local Go router + desktop app that sits between all your AI
clients and your MCP servers. You configure MCPs once in the Hub UI, then run
one command per client to connect it. That's it.
**The part I'm most proud of:** tool definition hash pinning. If an MCP you've
installed changes its tool descriptions or input schema silently (supply chain
attack territory), 1mcp blocks that tool and flags it for re-approval. You see
exactly what changed before anything runs.
**Stats:**
- 1ms warm calls / ~741ms cold start
- <30MB RAM footprint
- 18+ curated MCPs in the marketplace, each with a signed SHA256 digest
- Supports: VS Code, Cursor, Claude Desktop, Claude Code, Windsurf, OpenCode
**Honest caveats:**
- v0.3.1, public beta — expect rough edges
- Curl one-liner installer not live yet, using direct downloads for now
- Discord community not open yet
Desktop downloads (Mac/Win/Linux): https://www.1mcp.in
GitHub (Apache 2.0, source available): https://github.com/SaiAvinashPatoju/1mcp.in
If you're running 5+ MCPs across multiple clients, I'd really like to know if
this solves the pain or if I've missed something obvious.