24 hours since our https://t.co/8pGiq6OqEs launch, and we're about to hit 400 users! We also crossed 1,000 tool calls on our very first full day live 🚀
Your agent shouldn't carry every tool from every MCP server.
MCP1 lists only its own tools on a larger gateway (a few hundred tokens), finds tools by search, hands over a server's instructions with its first results at no extra turn, and cuts oversized answers to your limit.
https://t.co/C74rcoDBzC
How we measured: a simulated agent asking the public DeepWiki server through old and new MCP1, 3 runs each, with database latency set to what we see in production. Startup context is measured on 10 public servers.
Details: https://t.co/MYGQZV8pTZ
We just made MCP1 faster. Measured before and after on the public DeepWiki server:
MCP1's time per session: 2.5 s → 0.9 s
Database round trips: 22 → 12
Tokens re-read per question: 22% fewer
Plus 12× less to load at startup than connecting 10 servers directly.
Here’s how:
Two smaller ones:
If an agent already knows a tool's name, it can call it without searching. A wrong guess answers with the closest matches, so guessing costs nothing.
Connections to upstream servers stay open while in use, and a search opens the one it found.
Here’s MCP1 in 2 minutes.
Install a server from the catalog, share it through a gateway, connect Claude Code with one URL, and see every tool call your team makes.
Free for individuals: https://t.co/BGbJFgGsGT
🚀 @ProductHunt just joined MCP1!
Your AI agents can now explore trending products, discover new launches, and keep up with what's happening on Product Hunt, all through MCP.
One gateway. More superpowers for your agents.
Try it out 👉 https://t.co/CEf15sipV2