nake.si MCP
# nake.si MCP
This is a public stdio MCP server for the adult AI image and video site at https://nake.si/.
It reads public HTML from that host only. It does not generate images or videos, and it does not read or write API keys.
## Tools
### get_site_overview
Fetches https://nake.si/ and returns the page title, the meta description, and up to 20 links that stay on `nake.si`.
### read_public_page
Reads one public page. `path` must start with `/`, for example `/`. Absolute URLs and protocol-relative URLs on any other host are rejected before a request is sent. `script` tags are removed. The returned text is at most 4000 characters.
## Run
Node.js 20 or newer:
```bash
npm install
node src/index.js
```
Example client config:
```json
{
"mcpServers": {
"nake-si": {
"command": "node",
"args": ["/absolute/path/to/nake-si-mcp/src/index.js"]
}
}
}
```
## Check
```bash
npm run smoke -- --reject https://example.invalid/
```
`--reject` must be a different host. The command fails if the homepage cannot be read or if that host is not rejected.
TDQS
Scored across 2 tools
get_site_overview and read_public_page overlap somewhat, since read_public_page with path '/' reaches nearly the same content, but the overview tool is distinguished by returning structured metadata (title, description, curated link list). An agent can reasonably tell which to pick: one for a quick site summary, one for fetching a specific page.
Both names are snake_case verb_noun constructions (get_site_overview, read_public_page), which is predictable and readable. Minor inconsistency in verb choice (get vs read) for what are conceptually both retrieval operations.
Two tools is thin for any server, though the stated purpose (read-only access to a single public site) is narrow enough that a large surface isn't warranted. The overview tool is arguably a convenience wrapper over the page reader, so the set feels borderline rather than complete.
Covers the basics of reading a public site, but has notable gaps: no way to discover pages beyond the 20 links on the homepage, no search or sitemap/listing operation, and no link-following or crawl capability. Agents hitting a page whose links are truncated will have no workaround.