# AI Hosting | xCloud

> AI hosting where your agents run, and where they run your servers. Deploy Claude Code, Codex, OpenClaw, Hermes, Paperclip, Odysseus or n8n onto a server you own, then run that server by telling your agent what you want. MCP, Agent Skills and the Public API are free on every xCloud account.

**Note for agents:** `/ai-hosting/` is a hub. It routes to the AI product pages rather than replacing them; each product's own page is canonical for its setup steps and pricing. The machine-readable setup script is [`/agent-setup/prompt.md`](/agent-setup/prompt.md), and live plan prices are at `https://app.xcloud.host/api/v1/catalog/pricing` (JSON).

**Free plan:** 1 server, 10 sites, and MCP, Agent Skills and the Public API included. No card.

## Works with the agent you already use

Setup guides, one per client: [Claude Code](/agents/claude-code/), [Claude](/agents/claude/), [Claude Cowork](/agents/claude-cowork/), [Cursor](/agents/cursor/), [Codex](/agents/codex/), [Windsurf](/agents/windsurf/), [OpenCode](/agents/opencode/), [Zed](/agents/zed/), [Kiro](/agents/kiro/), [GitHub Copilot](/agents/github-copilot/), [Grok Bot](/agents/grok-bot/), [Hermes Agent](/agents/hermes-agent/). Anything that speaks Streamable HTTP MCP connects too. All guides: [/agents/](/agents/).

## One account, two directions

Your agent drives xCloud. xCloud runs your agents.

| Direction | What it means | Cost |
| --- | --- | --- |
| Agents **drive** xCloud | The agent you already use reaches in and operates your infrastructure, through MCP, the Agent Skills or the Public API | Free with every account, including the free plan |
| xCloud **runs** agents | The agents you deploy live on servers you own | Server cost only; see the table below |

What the skills cover, by area:

- **Deploy** (`xcloud:deploy`): Git, Docker and one-click deploys, with a dry run.
- **Troubleshoot** (`xcloud:troubleshoot`): why a site errors: status, events, logs.
- **Back up** (`xcloud:sites`): backups, restores and staging, on request.
- **Secure** (`xcloud:ssl`, `xcloud:servers`): certificates, firewall, fail2ban, updates.

## Set it up your way

**Do it yourself**

1. Create a free xCloud account (`https://app.xcloud.host/register`). No card needed.
2. Go to Settings → Developers → MCP and copy the URL.
3. Paste it into your agent, then approve the sign-in in your browser.

```
claude mcp add xcloud --transport http https://app.xcloud.host/mcp
```

**Let your agent do it (recommended)**

Paste one line into Claude Code, Codex or Cursor:

```
Set up xCloud by following https://xcloud.host/agent-setup/prompt.md
```

It reads the setup file, connects the MCP server and installs the skills, and stops only for the two steps a human has to do: the browser sign-in and a restart.

## Pick how your agent gets in

Three doors, same account. All three are free with every xCloud account, including the free plan, and reach the same operations. They differ in what you set up and in what holds the credentials. Every command, endpoint, scope and response below is verbatim from [`/agent-setup/prompt.md`](/agent-setup/prompt.md).

### Door 1: xCloud MCP (recommended) · [/mcp/](/mcp/)

One endpoint for Claude Code, Claude Desktop, Cursor, Codex, Windsurf, OpenCode or anything else that speaks MCP. OAuth fires the first time the agent reaches for an xCloud tool, so no token sits in a config file, and you revoke access from the dashboard rather than by editing a machine.

```
claude mcp add xcloud --transport http https://app.xcloud.host/mcp
codex  mcp add xcloud --url https://app.xcloud.host/mcp
```

Anything that takes an `mcpServers` block:

```json
{ "mcpServers": { "xcloud": { "url": "https://app.xcloud.host/mcp" } } }
```

- Stdio-only clients connect through `npx mcp-remote https://app.xcloud.host/mcp`.
- Headless and CI use a token scoped to `mcp:invoke`, since no browser can finish OAuth.

### Door 2: xCloud AI Agent Skills · [/ai-agent-skills/](/ai-agent-skills/)

The MCP hands your agent the tools; the skills teach it xCloud's workflows: which tool to reach for, the guardrails, how to poll an asynchronous job to completion, and how to chain a multi-step change. Nine skills ship in one plugin and the agent picks; you never name one. They also ship as an Agent Plugins package for ChatGPT, Codex, Cursor, GitHub Copilot, Kiro and VS Code.

| Skill | Owns |
| --- | --- |
| `xcloud:deploy` | Git, Docker and one-click deploys, staging, new WordPress sites |
| `xcloud:troubleshoot` | Why a site errors: status, events, logs, services |
| `xcloud:performance` | Why a site is slow: cache layers, PageSpeed, PHP |
| `xcloud:servers` | PHP, services, cron, firewall, fail2ban, reboots |
| `xcloud:sites` | Backups, domains, cache, SSH, git settings, staging |
| `xcloud:wordpress` | Plugins, themes, updates, magic login, vulnerabilities |
| `xcloud:ssl` | Certificates: view, install, renew, delete |
| `xcloud:billing` | Plan, invoices, subscriptions, add-ons |
| `xcloud:account` | Teams, alerts, API tokens, integrations, blueprints |

```
claude plugin marketplace add xCloudDev/xcloud-agent-skills
claude plugin install xcloud@xcloud-agent-skills
```

Then `/reload-plugins` to activate.

### Door 3: xCloud Public API

Everything the MCP exposes is a Public API operation underneath, so there is no capability gap between the paths. Use it directly when the integration lives inside your own product: a backend service, a provisioning script, a CI job, an agent loop of your own.

- **Base URL:** `https://app.xcloud.host/api/v1`
- **Auth:** Bearer token (Laravel Sanctum personal access token)
- **Reference:** OpenAPI 3.0.3 at `https://app.xcloud.host/api/v1/docs`

```bash
curl -s https://app.xcloud.host/api/v1/health
# -> {"status":"ok","version":"v1"}
```

`/health` needs no auth, which is what makes it useful: it isolates a network problem from a credential one.

**Token scopes:** `Full Access`, `read:servers`, `write:servers`, `read:sites`, `write:sites`, `read:billing`, `read:addons`, `write:addons`, `mcp:invoke`. Give each token the narrowest scope that covers its job.

⚠️ **Write operations are asynchronous.** They return a task you poll, not a finished resource, so build in polling and a timeout.

## Every change is approved, tracked and verified

The control plane (MCP, Agent Skills and the Public API) is **free** on every account.

- **No token sits in a config file.** Sign in with OAuth in the browser. Access is scoped to your team and you can revoke it from the dashboard at any time.
- **Narrowest scope wins.** Give each token only what its job needs.
- **Risky changes ask first.** Destructive tools ask for confirmation before they run, whichever client you connect from.
- **Backup first, then change.** The skills back up before a bigger change and wait for the job to finish, then point you to the one-click restore.

## Things people actually type

Every one of these is inside the documented scope of the nine skills:

- "Which of my sites have plugin vulnerabilities?" → `xcloud:wordpress`
- "Reboot staging-01 and tell me when it's back." → `xcloud:servers`
- "Renew the certificate on every domain expiring this month." → `xcloud:ssl`
- "What PHP version is each server on?" → `xcloud:servers`
- "Purge the cache on everything in production." → `xcloud:sites`
- "Back up the shop before I touch the theme." → `xcloud:sites`

## Live in three steps

1. **Pick what to run.** An agent from the table below, or your own stack from Git or a Docker Compose file.
2. **Connect your agent.** One MCP URL works for every client.
3. **Ask.** "Build me a site." "Why is this slow?" "Back it up."

## The agents: start from what you want to do

Pages: [Claude Code hosting](/claude-code-hosting/) · [Codex hosting](/codex-hosting/) · [OpenClaw hosting](/openclaw-hosting/) · [Hermes Agent hosting](/hermes-agent-hosting/) · [Paperclip hosting](/paperclip-hosting/) · [Odysseus hosting](/odysseus-hosting/) · [DeepSeek Harness hosting](/deepseek-harness/) · [n8n hosting](/n8n-hosting/) · [Ollama hosting](/ollama-hosting/) · [LibreChat hosting](/librechat-hosting/) · [AnythingLLM hosting](/anythingllm-hosting/) · [AI agent hosting](/ai-agent-hosting/) · [AI VPS](/ai-vps/) · [MCP server](/mcp/) · [AI Agent Skills](/ai-agent-skills/)

| I want… | Deploy | How you reach it | From |
| --- | --- | --- | --- |
| a coding agent that keeps working after I close the laptop | [Claude Code](/claude-code-hosting/) | Terminal, Claude web and mobile apps | $5.00/mo |
| to put the app Codex built on a real domain | [Codex](/codex-hosting/) | Your own domain, redeployed on git push | $5.00/mo |
| an open-source coding agent I can extend with plugins | [DeepSeek Harness](/deepseek-harness/) | A managed Cloud VPS, your choice of models and tools | $9.99 first month, then $19.99/mo |
| an assistant I can text like a person | [OpenClaw](/openclaw-hosting/) | Telegram, set up for you | $9.99 first month, then $19.99/mo |
| an agent that remembers everything between conversations | [Hermes](/hermes-agent-hosting/) | Telegram, plus your tools, files and APIs | $9.99 first month, then $19.99/mo |
| a workforce of agents with roles and budgets | [Paperclip](/paperclip-hosting/) | Its own dashboard | $9.99 first month, then $19.99/mo |
| a private AI workspace: chat, memory, research, documents | [Odysseus](/odysseus-hosting/) | A private web workspace from a Docker Compose file, your own LLM keys | $5.00/mo |
| to automate the plumbing between all of the above | [n8n](/n8n-hosting/) | 500+ integrations, webhooks, AI nodes. Needs 4 GB RAM | $24.00/mo |
| to run something that is not on this list | [Your own stack](/self-managed-hosting/) | Deploy from Git, or a Docker Compose file | $3.00/server/mo |

**Prices.** Claude Code, Codex and Odysseus run on a standard managed server, from $5.00/mo. n8n needs at least 4 GB of RAM and 2 CPU cores, so it starts on the $24.00/mo plan. OpenClaw, Hermes, Paperclip and DeepSeek Harness run on a Cloud VPS sized for an always-on agent: $9.99 for the first month, then $19.99/mo. See [/pricing/](/pricing/).

### What an agent server is

OpenClaw, Hermes, Paperclip and DeepSeek Harness run on a Cloud VPS built for one always-on agent:

- Each server runs one agent and nothing else xCloud manages: no extra sites, Docker apps or one-click apps.
- There is no GPU. Your model runs at its provider.
- Backups are not included on the AI-agent plans.
- Hosting does not include AI credits. Bring your own key from Anthropic, OpenAI, OpenRouter, Moonshot AI (Kimi) or Gemini.

## Every server comes ready

- **Hardware:** 3GHz+ Intel Xeon · NVMe SSD · NGINX or OpenLiteSpeed · 31 locations
- **Security:** free SSL, auto-renewed · firewall + fail2ban · automated patching · AI Bot Blocker (configurable, so you choose whether crawlers like GPTBot can read your sites)
- **Recovery (standard managed servers):** scheduled backups · any S3-compatible store · snapshots · one-click restore. AI-agent plans do not include backups.
- **Visibility:** server and site monitoring · access and deploy logs · vulnerability scans · PageSpeed history

## What people running agents on xCloud say

> "…I reluctantly switched from RunCloud after my Server specialist showed me a lifetime deal…I took it. And have been super happy with all that you offer. With over 160 websites now, I have my AI (Clawbot) scanning and reviewing my servers daily."
>
> David Oralevich, Founder, Designs By Dave O. [Read the case study](/case-studies/designs-by-dave-o/)

## What's new in AI on xCloud

- **21 Sep 2026, [v2.8.8](/changelog/v2-8-8/):** a more capable MCP and Public API. Agents find the right operation, validate inputs before running, and recover failed deployments. Plus multi-team API access.
- **17 Aug 2026, [v2.8.1](/changelog/v2-8-1/):** managed DeepSeek Harness hosting.
- **11 Aug 2026, [v2.7.9](/changelog/v2-7-9/):** 59+ new one-click apps.
- **1 Jul 2026, [v2.6.9](/changelog/v2-6-9/):** xCloud AI Agent Skills launch.

## Read before you pick an agent

- [OpenClaw vs Paperclip vs Hermes (2026)](/openclaw-vs-paperclip-vs-hermes/)
- [Hermes Agent vs OpenClaw: which AI agent to host in 2026?](/hermes-agent-vs-openclaw/)
- [OpenClaw + n8n integration: secure AI agent workflows](/openclaw-n8n-integration/)
- [5 OpenClaw sub-agent configurations](/openclaw-sub-agent-configurations/)
- [Best AI agent hosting in 2026](/best-ai-agent-hosting/)
- [xCloud vs Railway](/xcloud-vs-railway/)

## FAQ

**What is free, and what do I pay for?** The control plane is free. MCP, the AI Agent Skills and the Public API are included with every xCloud account, including the free plan. You pay only when you run a server. Your AI assistant's model usage is billed by its own provider.

**Should I use MCP or the Agent Skills?** MCP first: one endpoint, every MCP-compatible client, no token to paste. The Skills add ready-made workflows on top: a Claude Code plugin, and the same nine skills as a portable Agent Plugins package for ChatGPT and Codex, Cursor, GitHub Copilot, Kiro and VS Code.

**Why do the prices in the table differ?** Because they are different products. Claude Code, Codex and Odysseus run on a standard managed server, from $5.00/mo; n8n needs 4 GB of RAM, so it starts on the $24.00/mo plan. OpenClaw, Hermes, Paperclip and DeepSeek Harness run on a Cloud VPS: $9.99 for the first month, then $19.99/mo.

**Can I bring my own model, and my own server?** Both. Anthropic, OpenAI, OpenRouter, Moonshot AI (Kimi) or Gemini today, with Grok, xAI and Mistral coming. Run everything on an xCloud Managed Server or on your own machine at DigitalOcean, Vultr, Linode, AWS or another provider.

**Does hosting include AI credits?** No. Bring your own key; model usage is billed by its provider.

**Can I run more than one agent, or a website, on an agent server?** No. Each AI-agent server runs one agent. Use a standard managed server for websites, Docker apps or one-click apps.

**Does OpenClaw work with WhatsApp?** Telegram is set up for you. WhatsApp is not bundled as a native channel.

**Where do I find my MCP URL?** Settings → Developers → MCP. The endpoint is `https://app.xcloud.host/mcp` for every client.

**Can my agent set itself up?** Yes. Ask it to follow `https://xcloud.host/agent-setup/prompt.md`. It stops only for the browser sign-in and a restart.

**What stops my agent from breaking something?** It acts only with the permissions on the token you issued, scoped to your team. Destructive tools ask for confirmation, the skills back up before a bigger change, and you can revoke access at any time.

## Start

- Create a free account: `https://app.xcloud.host/register`
- Let your agent set it up: [`/agent-setup/prompt.md`](/agent-setup/prompt.md)
- Setup guides for every agent: [/agents/](/agents/)
- Browse Agent Skills: [/ai-agent-skills/](/ai-agent-skills/)
- Whole-site catalogue: [/llms.txt](/llms.txt)