MCP serializes all tool schemas into every conversation turn. With 50 tools, that's 121,000 tokens/turn burned on definitions alone — before the model reads a single user message.
NEKTE is an open protocol with 8 primitives designed for token efficiency:
MCP NEKTE Savings
─────────────────────────────────
5 tools 3,025 345 -89%
30 tools 54,450 1,155 -98%
100 tools 302,500 2,325 -99%
Lazy Discovery · Zero-Schema Invocation · Token Budgets · Streaming · Task Lifecycle
discover → invoke → delegate
↓ ↓ ↓
L0/L1/L2 hash-based SSE/gRPC streaming
~8 tok 0 extra with cancel/suspend/resume
| Package | Description | |
|---|---|---|
| Core | @nekte/core |
Types, schemas, hashing, budget, codec, task state machine |
| Client | @nekte/client |
HTTP/gRPC transports, discovery cache, streaming + cancel |
| Server | @nekte/server |
Capability registry, task registry (DDD), auth, SSE + gRPC |
| Bridge | @nekte/bridge |
MCP proxy — 90%+ savings, zero backend changes |
| CLI | @nekte/cli |
discover, invoke, health, card, bench |
| Python | nekte-python |
Python SDK |
# TypeScript
pnpm add @nekte/client @nekte/server# Python
pip install nekte# Run the demo
git clone https://github.com/nekte-protocol/nekte.git
cd nekte && pnpm install && pnpm demoNEKTE complements existing protocols — it doesn't replace them:
- + MCP — MCP connects agents to tools. NEKTE connects agents to agents. The bridge enables instant adoption.
- + A2A — A2A prioritizes enterprise governance. NEKTE prioritizes token efficiency.