MEW Protocol (pronounced like a cat's "mew" 🐾) keeps humans in the loop - you participate directly in agent workspaces, see every operation, and maintain control through gateway-enforced capabilities.
MCP: human ──► agent ════► mcp_server
↑
trust boundary
(agent decides)
A2A: human ──► agent₁ ════► agent₂
↑
trust boundary
(agent₁ decides)
MEW: human ──►┌─────────┐◄── agent
│ GATEWAY │
mcp_server ──►└─────────┘◄── agent₂
↑
trust boundary
(gateway enforces)
Legend: ──► trusted flow, ════► potentially untrusted
In MEW, humans aren't observers - they're active participants alongside agents and MCP servers:
- 👀 See everything: Every message, every operation, nothing hidden
- 🎛️ Control access: You grant and revoke capabilities, not the agents
- ✅ Review proposals: Agents propose, you decide what executes
- 💬 Direct interaction: Chat with any participant, send MCP requests directly
The proposal mechanism ensures humans stay in control. New agents start with proposal-only access - they suggest operations but can't execute until you approve. As you observe their behavior, you grant specific capabilities. Trust is earned, not assumed.
- You define the rules: Set initial capabilities for each participant
- Agents propose: Limited agents suggest operations via proposals
- You review: Approve, reject, or delegate review to trusted participants
- You grant trust: Expand capabilities for agents that prove reliable
- You maintain control: Revoke capabilities instantly if needed
v0.3 - Released 2025-01-09 🎉
MEW Protocol is in experimental phase (v0.x) with breaking changes allowed between versions. See spec/v0.3/SPEC.md for the current specification.
npm install -g @mew-protocol/cli# Create a new directory for your workspace
mkdir my-workspace
cd my-workspace
# Start MEW - it will guide you through setup
mew
# 🐱 MEW Protocol workspace starting...
# 🎯 Setting up your workspace...
# 🤖 Choose your AI agent (Claude, GPT-4, etc.)
# 📁 Select tools (filesystem, web search, etc.)
# 🔐 Configure capabilities (proposals vs direct execution)
#
# Type 'help' for commands or start chatting!That's it! MEW guides you through setting up your workspace with:
- 🧑💻 You participating directly via terminal
- 🤖 AI agent(s) with configured capabilities
- 🛠️ Tools and MCP servers for extended functionality
- ✅ Proposal/approval flow for safe operations
- 🎓 Progressive trust as patterns prove safe
- 📋 Current Specification (v0.3)
- 📝 Draft Specification (next version)
- 🏗️ Architecture Decision Records
- 📜 Changelog
The name playfully evokes "herding cats" 🐈⬛🐈🐈⬛ - the quintessential challenge of coordinating multiple independent, autonomous agents. MEW Protocol provides the framework to bring order to this chaos, teaching the "cats" to work together effectively in a shared workspace. 🌠