Pre-Submission Checklist

• I have verified this wouldn’t be more appropriate as a feature request in a specific repo
• I have searched existing discussions to avoid duplicates

Introduction

Hey all! Fabri here (Product Manager at XMTP). Would love your feedback and explore collaboration opportunities around secure AI agent communications.

Overview

We’re seeing increased demand for “agents on XMTP,” which led us to build agent-starter—a minimal library for end-to-end encrypted (E2EE) agent messaging. It works with various AI frameworks or even web2 APIs and pairs nicely with xmtp.chat for a complete dev experience.

Reference: xmtp-agents repo

Why XMTP?

• End-to-end & compliant: Servers and clients only see ciphertext, meeting strict security and regulatory standards.
• Open-source & trustless: Built on the MLS protocol, it removes centralized certificate authorities with cryptographic proofs.
• Privacy & metadata protection: Offers anonymous or pseudonymous usage with no tracking of timestamps, routes, IPs, or device info.
• Decentralized: Operates on a peer-to-peer network, eliminating single points of failure.
• Groups: Allows multi-agent (or multi-human) group chats with access control and secure collaboration.

Why XMTP + MCP?

MCP (Model Context Protocol) is like a “USB-C for AI,” standardizing how LLMs connect to data and tools. XMTP can add an E2EE layer, the "cable", protecting agent communications and metadata:

1. Standardized security: MCP sets the interface; XMTP secures data in transit and at rest.
2. Compliance: Ideal for finance, healthcare, legal, etc.
3. Scalability: Plug-and-play with existing frameworks, web2/web3 systems, token-gated access.

Scope

• Protocol Specification
• SDK Features
• Documentation
• Developer Experience
• Other

Looking for:

• Your thoughts on integrating XMTP’s E2EE with MCP
• Any security/architecture considerations from the MCP side
• Potential next steps for a proof-of-concept or pilot

Thanks in advance for any feedback or suggestions!