What is MCP
MCP: the protocol that makes AI clients interchangeable.
The Model Context Protocol (MCP) is an open standard that defines how AI assistants communicate with external tools and services. An AI client that supports MCP can call any MCP server's tools — without custom code or vendor-specific integrations. DetachDev implements an MCP server that exposes your Claude Code® sessions as tools, making them accessible from any compliant client.
Open standard
MCP is not proprietary to any AI vendor. Claude.ai®, ChatGPT®, and any other MCP-capable client can connect to DetachDev without custom integration work.
Tool-based interface
DetachDev exposes session operations as MCP tools: list sessions, resume session, send prompt, approve permission, get history. AI clients discover these tools automatically from the MCP manifest.
HTTPS transport
All MCP traffic runs over standard HTTPS. No custom protocols, no special firewall rules, no VPN required. Your MCP endpoint is authenticated per request using your session token.
How it works
Outbound-only architecture. No open ports.
The most important architectural decision in DetachDev is that your dev machine never accepts inbound connections. Instead, the daemon on your machine connects outbound to the DetachDev platform, which acts as a relay between your AI client and your sessions.
Your AI client sends an MCP request to the DetachDev platform over HTTPS. The platform relays it to your dev machine over the persistent outbound WebSocket the daemon established at startup. The daemon forwards it to the Claude Code® session process and streams the response back the same way. At no point does anything initiate a connection to your machine.
What crosses the wire
Commands and output only. Never source code.
DetachDev is a relay, not a storage system. The distinction matters for your security posture.
What is relayed
Prompts you send to Claude Code®. Output and tool call results Claude Code® returns. Permission prompts and your responses to them. Session metadata: name, status, last modified.
What never leaves your machine
Your source code files. Your file system structure. Environment variables and secrets. Your git history. Anything Claude Code® reads from disk but does not include in its output response.
For the full security architecture including encryption, audit logging, and enterprise controls, see the security page.
MCP endpoint
One endpoint. Every session. Any client.
Once registered, your personal MCP endpoint is available immediately. Add it to any MCP-compatible AI client once — it gives that client access to all sessions on all your registered devices.
endpoint: https://mcp.detachdev.io/u/your-token
# MCP protocol version
protocol: MCP 2024-11-05
# Compatible with
clients: Claude.ai, ChatGPT, any MCP/HTTP client
Questions
Frequently Asked Questions
What is MCP and how does it enable remote development?
MCP (Model Context Protocol) is an open standard for AI clients to communicate with tools and services via a structured HTTP interface. DetachDev implements an MCP server that exposes your Claude Code® sessions as tools — so any compliant AI client can list, resume, and control sessions remotely without SSH or custom integrations.
Does DetachDev require open inbound ports on my dev machine?
No. The DetachDev client daemon makes an outbound WebSocket connection to the DetachDev platform. The platform then relays MCP requests from your AI client to the daemon. No inbound ports are opened on your dev machine at any point.
Which MCP version does DetachDev implement?
DetachDev implements MCP protocol version 2024-11-05 over HTTPS. This is the current stable specification and is compatible with all major MCP-capable AI clients.
How is the MCP endpoint authenticated?
Each request to your MCP endpoint is authenticated using your session token, which is embedded in the endpoint URL. You can rotate or revoke tokens from the dashboard at any time. Enterprise plans support additional authentication layers including IP allowlisting.