MCP server
Connect AI tools to your published docs through a Model Context Protocol server.
Every Bloques site exposes a Model Context Protocol (MCP) server — a read-only endpoint that lets AI tools search and read your docs while they work. The server is active by default on every site, on your <subdomain>.bloques.site URL and any custom domain. No setup required.
The MCP server differs from the LLM endpoints: instead of pasting a URL for one-time context, an AI client connects once and then searches and fetches pages on demand.
Server URL
Point your AI tool at this URL:
https://<subdomain>.bloques.site/mcpThe endpoint uses the streamable HTTP transport and works on custom domains too — replace the host with your domain.
Connect an AI tool
Add the server URL to your tool's MCP configuration. Replace your-site.bloques.site with your site's host, and rename acme-docs to whatever you want the connection labeled.
Cursor
Add the server to .cursor/mcp.json:
{
"mcpServers": {
"acme-docs": {
"url": "https://your-site.bloques.site/mcp"
}
}
}VS Code
Add the server to .vscode/mcp.json:
{
"servers": {
"acme-docs": {
"type": "http",
"url": "https://your-site.bloques.site/mcp"
}
}
}Claude Code
Run the claude mcp add command:
claude mcp add --transport http acme-docs https://your-site.bloques.site/mcpOther clients
Any client that supports MCP over HTTP — Claude Desktop, Windsurf, and others — works the same way. Open the client's MCP or connectors settings and add the server URL. The server reports its name as your domain followed by docs, for example your-site.bloques.site docs.
Clients without HTTP support
Some clients connect only over standard input/output (stdio) instead of HTTP. Bridge them to the endpoint with mcp-remote:
{
"mcpServers": {
"acme-docs": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://your-site.bloques.site/mcp"]
}
}
}Available tools
An AI client calls these tools automatically while it answers — you don't invoke them by hand.
| Tool | Parameters | Returns |
|---|---|---|
search_docs | query, language (optional), limit (optional, 1–20, default 10) | The most relevant pages and sections, each with a snippet and URL. |
list_pages | language (optional) | Every page on the site as a markdown link (title and URL). |
get_page | path (empty for the home page), language (optional) | One page as raw markdown. |
The language parameter accepts a language code your site publishes (en or es). It defaults to en and falls back to the default language when a translation is missing. See Internationalization.
What the server exposes
The server is read-only and returns only your published content — the same pages your public site shows. Drafts and unpublished changes stay private, and the server can't modify your docs or your repository.