In 1999, Linux seemed hopeless against Windows. Today, it runs 100% of supercomputers and 96% of web servers.
Linux didn't win by fighting Windows. It won by enabling everything else.
I'm applying the same strategy to AI agent communication with MXP (Mesh eXchange Protocol).
The Problem
AI agents are evolving from isolated assistants to interconnected systems. But they're stuck using HTTP/JSON—a protocol designed for humans clicking buttons, not machines coordinating at microsecond scale.
The overhead matters:
HTTP/JSON: 2,262ns to encode a 256-byte message
MXP: 60ns for the same message
That's 37x faster. Verified. Reproducible.
The Solution: Be the Linux, Not the Windows
Google launched A2A (Agent-to-Agent Protocol) with 50+ partners. Fighting that would be like fighting Windows in 1999.
Instead, MXP takes the Linux approach:
A2A-compatible: Same semantics, faster transport
Open source: MIT/Apache-2.0, public domain spec
Enabling, not competing: Make other projects faster
use mxp::{Message, MessageType};
// Same agent logic, 37x faster encoding
let msg = Message::new(MessageType::Call, b"Hello, agent!");
let bytes = msg.encode(); // 60ns, not 2,262ns
Why It Matters
When agents coordinate 1000x per second (heartbeats, discovery, token streaming), protocol overhead becomes real CPU time.
Scenario HTTP/JSON MXP
1M messages 2.26 seconds CPU 0.06 seconds CPU
10 agents × 100 msg/s 226ms/s 6ms/s
Try It
cargo add mxp
Or explore the repo: github.com/yafatek/mxp-protocol
Star if you believe open infrastructure wins. ⭐
The "enabler not competitor" positioning
What SDKs to prioritize (JS is next, then Python)
Whether A2A compatibility matters to you
Repo: https://github.com/yafatek/mxp-protocol
Top comments (0)