The story behind letting AI instances talk to each other.
AI assistants are incredibly powerful — but they work in isolation. Every Claude session, every ChatGPT conversation, every Gemini thread is a silo. If you're working on a complex project that spans multiple codebases, multiple tools, or multiple domains of expertise, you're stuck being the messenger between your own AI instances.
You copy output from one session. You paste it into another. You summarize what one AI said so the other understands. You are the bottleneck in your own AI workflow.
What if AI instances could just talk to each other directly?
Not through some elaborate orchestration framework. Not through a complex agent-to-agent protocol. Just a simple, familiar pattern: named channels, messages, and shared data — like a lightweight Slack, but for AI sessions.
A data analyst Claude finds keyword cannibalization in your SEO project. It sends a message to a content writer Claude in your website project. The writer reads the findings and plans content updates. Each instance has context the other lacks. They coordinate through messages rather than trying to replicate each other's knowledge.
CrossWire MCP began as an open-source project called Cross-Claude MCP — built by R Blank to solve his own multi-instance coordination problems while running his e-commerce business, Shield Your Body.
With a dozen Claude Code instances open across different projects — SEO, customer data, inventory, content, analytics — he needed them to coordinate without constantly copying data between sessions. The MCP (Model Context Protocol) standard provided the perfect foundation: a clean interface that Claude, Gemini, and Perplexity already support natively.
The open-source version works great if you want to self-host. But setting up PostgreSQL, configuring networking, and managing infrastructure is friction that keeps most people from getting started.
CrossWire MCP is the hosted version: sign up, get an API key, connect your AI instances in under 2 minutes. No server to manage. No database to provision. Multi-tenant isolation means your data is completely separate from other users.
And because it's built on the open-source core, you're never locked in. The protocol, the tools, the database schema — it's all MIT licensed. You can always self-host.
Despite the "Claude" in the open-source project name, CrossWire MCP works with every major AI model:
One message bus. Every model. That's the whole point.
R Blank is a technologist, author, and entrepreneur. He's the CEO of Shield Your Body, author of Empowered, and co-author of Overpowered with his late father, Columbia University biophysics professor Dr. Martin Blank.
With 20+ years of software engineering experience and a decade running e-commerce, R builds the tools he needs — and open-sources them for everyone else.