Connect Claude, ChatGPT & AI Agents

What is MCP

The Model Context Protocol is an open standard that lets AI assistants call real tools. Hostwares runs an official MCP server, so an assistant can deploy a repo, read build logs, set environment variables and attach domains for you — in plain language, without leaving your editor.

It is hosted by us at https://mcp.hostwares.com — nothing to install, and it updates as the platform does.

Quickstart (Claude Code)

Create an agent key at Dashboard → API Keys (pick “Deploy my apps”), then run:

claude mcp add --transport http hostwares https://mcp.hostwares.com \
  --header "Authorization: Bearer hwk_your_key_here"

That is the whole setup. Now just ask:

> Deploy this repo to Hostwares as my-app
> Why is my-app returning 502? Check the logs
> Set DATABASE_URL on my-app
> Point app.example.com at my-app

Verify it connected with claude mcp list.

Claude.ai, Claude Desktop & ChatGPT

These apps connect with your Hostwares login. No key to copy. Add a custom connector with this URL:

https://mcp.hostwares.com

• Claude.ai / Claude Desktop: Settings → Connectors → Add custom connector → paste the URL → Connect.
• ChatGPT: Settings → Connectors → Create → paste the URL, authentication OAuth.
• You’re sent to hostwares.com to sign in and choose what the app may do. Destructive access starts unticked.
• Disconnect any time under Dashboard → API Keys → Connected apps.

Other clients

Cursor, VS Code, Windsurf, Lovable

Each client keeps the URL and the key in its own MCP config, and the field names differ a little. The connect guide has the exact file and snippet for each one. This is the Cursor form (~/.cursor/mcp.json); Claude Desktop connects through Settings → Connectors above, not its config file:

{
  "mcpServers": {
    "hostwares": {
      "url": "https://mcp.hostwares.com",
      "headers": {
        "Authorization": "Bearer hwk_your_key_here"
      }
    }
  }
}

ChatGPT & other remote-MCP clients

Add a custom MCP connector pointing at https://mcp.hostwares.com with the header Authorization: Bearer hwk_.... Any client that speaks Streamable HTTP works.

Local stdio (clients without remote MCP)

A few clients only speak local stdio. For those, run our bridge — it forwards to the same hosted server, so you get the same tools without them going stale:

{
  "mcpServers": {
    "hostwares": {
      "command": "npx",
      "args": ["-y", "hostwares-mcp"],
      "env": { "HOSTWARES_API_KEY": "hwk_your_key_here" }
    }
  }
}

Check it works

A plain GET returns the server descriptor, so you can sanity-check reachability:

curl https://mcp.hostwares.com

Tools

These call the same internal services as the dashboard and the hw CLI, so results are identical whichever way you drive the platform.

How deploys are billed

Call deploy without packId first and it returns the available resource packs so you can pick one. Databases currently require an active plan — see Plans.

Security

• Agent keys (hwk_) are scoped. You choose what each key may do (read only, deploy, databases, DNS, orders, destructive actions), when it expires, and a monthly order limit. An agent only sees the tools its key allows.
• Coding-agent presets. “Coding agent” = read + deploy + read your app’s database. “Coding agent + database’ adds database writes and a command-on-your-instance terminal — each change asks for your approval. These keys never get DNS, billing or destructive access. Call whoami for the exact scope→tool list.
• Nothing costs money or goes offline without you. Orders, new databases and destructive actions (stop a site, replace a domain, change secret env vars) return an approval link. You open it signed in to hostwares.com and approve; exactly what you saw is what runs. This holds even if your agent runs with permissions skipped.
• Every tool is scoped to the key’s owner — a key can only ever see and change that account’s resources. Our server passwords, internal IPs and infrastructure details are never returned by any tool.
• Every call is logged (tool, key, result — never your arguments’ values). See it under Dashboard → API Keys → Recent agent activity. Revoke a key there instantly.
• Older full-access sk_ keys still work with MCP until 3 January 2027, with approvals applied. Switch to an agent key before then.

Troubleshooting