Connect your agent over MCP
Mint a token, paste one block of configuration, and your coding assistant can create sites, deploy them, attach domains and read the status without you opening a dashboard.
Last updated on
What MCP is, in one paragraph
MCP is an open standard that lets an AI assistant call tools rather than just talk about them. We expose the entire feature set of staticsites.eu over it, because that is how the product was built rather than something added afterwards. An agent connected this way can take a project from a repository to a live URL without you opening the dashboard once.
1. Mint a token
- Open the dashboard and go to Settings, then API Tokens.
- Create a token. The value is shown once, at creation. Copy it now.
Treat it like a password. Anyone holding it can create and destroy sites in your workspace.
2. Paste the configuration
Claude Code, Claude Desktop and Cursor
Add this to your MCP configuration, replacing the token:
{
"mcpServers": {
"staticsites": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://app.staticsites.eu/api/mcp",
"--header",
"Authorization: Bearer YOUR_TOKEN"
]
}
}
}
Restart the client afterwards. mcp-remote bridges the stdio transport these clients expect to our
HTTP endpoint.
VS Code
Create or edit .vscode/mcp.json in your workspace, with the same block under a servers key
instead of mcpServers. VS Code picks it up automatically.
Anything else
The endpoint speaks the HTTP Streamable MCP transport. Send a POST to
https://app.staticsites.eu/api/mcp with your token in the Authorization header as a bearer token.
3. Check it worked
Ask your agent to call ping. It validates the connection and the token end to end and does nothing
else. If it comes back, you are connected.
Then ask it to call deploy_requirements. Rather than failing at you, it returns a checklist of
what is still missing, which on a fresh account is usually the GitHub App. That is the one thing an
agent cannot install for you.
The tools it gets
| Tool | What it does |
|---|---|
ping |
Check the connection and the token |
deploy_requirements |
Ask what is missing before creating a site |
validate_repo |
Confirm a repository exists and is readable |
provision_site |
Create a site from a repository and run the first build |
upload_site |
Create or update a site from a folder of built files |
deployment_status |
Read where a deploy has got to, and the address once live |
redeploy_site |
Send the last upload out again |
rollback_site |
Put an earlier deploy back on the live site |
list_deployments |
Read the deploy history |
list_projects |
Read every site in a workspace |
teardown_site |
Destroy a site and its hosting, irreversibly |
add_domain |
Attach a subdomain and get the record to create |
list_domains |
Read domains, DNS state and certificate state |
remove_domain |
Detach a domain |
recheck_domain |
Check DNS and the certificate now rather than later |
Two things only a human can do
An agent cannot install a GitHub App on your account or your organisation, and it cannot add a record at your DNS provider. Those are the two steps we ask you for, and we tell you exactly what to click when the moment comes. Everything else it can handle on its own.
FAQ
The questions worth answering before you hand us anything, including the ones where the answer is no.