Claude Code MCP: Deploy From the Chat
Stand up and orchestrate the whole Agentic Engineering Infrastructure straight from Claude Code over the Model Context Protocol. Answer a few questions, one tool call does the rest — zero server management.
The Model Context Protocol lets an AI assistant call real tools, and the Fractera connector is not tied to one vendor: it works with any MCP-capable code generator — Claude Code, OpenAI Codex, Gemini CLI, Qwen Code and Kimi Code, each on your own subscription. Point any of them at a private platform for agentic engineering: you describe what you want in plain language, and the agent provisions the same production stack you would get from the install form — without you touching a server console.
Deploy without leaving the chat
One tool call does the whole install
Add the connector and the agent asks for the essentials — your email, the VPS address and its password — then calls one tool that registers you, wipes the target machine and launches the IP-first install. You watch the steps stream back into the conversation in real time; there is no terminal, no manual setup, and nothing to copy and paste.
IP-first result: http://<your-ip>:3002
The deploy is IP-first, exactly like the server deploy. When it completes — typically in about ten minutes — your Admin workspace is reachable on plain HTTP at your server’s IP on port 3002. There is no domain and no HTTPS yet, and that is by design. Attaching your own domain with HTTPS is a separate, optional step you do later from inside the workspace.
Why the connector is more than a deploy button
The site is a knowledge base the agent can read
The connector exposes the project’s own knowledge base, so Claude can answer questions about Fractera while it deploys — what it is, what comes up on the server, how the pieces fit together. Read the same material on the project knowledge base page or explore the architecture visually.
Recovery and status are built in
A deploy is never a dead end. The connector can check the status of an in-flight install, return the finished address, and recover a failed run from a server token — so a transient hiccup is something Claude can pick back up instead of forcing you to start over.
“A private engineering stack on a server you own, stood up by an assistant that talks to you in plain language and never asks you to open a terminal.”
Prefer to deploy from the chat? Add the Fractera MCP connector and ask your agent — Claude Code, Codex, Gemini CLI, Qwen Code or Kimi Code — to deploy. Prefer a form? Use the install form instead — both reach the same place.
MCP · AI Agents
Complete Server Management via Claude Code MCP Server
Building and managing a production server through an AI agent inside your chat has never been this seamless. Connect Claude, Codex, or Gemini to the Fractera MCP server — deploy infrastructure, monitor installation, and launch new environments without leaving your conversation. You can also use the MCP server to diagnose and resolve any deployment issues directly from your AI chat.
Fractera MCP server URL
https://www.fractera.ai/api/mcp* Never used MCP before? Just copy the URL above and ask your AI agent (Claude, Codex, Gemini): "please connect the MCP server at this URL." The AI will walk you through what to click inside its own interface — setup takes no more than 15 seconds. Once connected, tell the agent: "deploy Fractera for me" or "my deploy failed, here is the server_token" — it takes over from there.
Full technical reference & security FAQ →Step-by-step: deploying via the MCP connector
A custom implementation for deploying a server that runs Claude Code only.
Experience is what you are left with when you do not get what you wanted… I am a very experienced man.
Roma ArmstrongFounder at Fractera.aiFrequently asked questions
- What is the Fractera MCP connector used for?
- It lets Claude (and any MCP-compatible assistant) deploy a private Fractera Agentic Engineering Infrastructure directly from the chat. It asks a few questions and calls a single tool that provisions a VPS, runs the install and streams real-time progress — no manual setup required.
- Does deploying over MCP need a custom domain or HTTPS?
- No. The deploy is IP-first: your workspace goes live on plain HTTP at http://<your-ip>:3002 in about ten minutes, with no DNS or certificate wait. Attaching your own domain with HTTPS is an optional later step done from inside the workspace.
- Is the deployed stack open-source and self-hosted?
- Yes. The layer that runs on your VPS is open-source and runs entirely on your own server — no data leaves it. You own the code, the database and the AI memory, whether you deploy over MCP or from the install form.