Docs MCP server
The same documentation you're reading is exposed over the Model Context Protocol so an agent (or an IDE assistant) can search it and pull full pages into its context — instead of guessing at the API or the SDK.
Connect it
It's a remote MCP server over Streamable HTTP — no local checkout, no install. It rides this
same domain (reverse-proxied at /mcp), so there's no extra host to trust. Point any MCP client
at it:
{
"mcpServers": {
"noolog-docs": { "type": "streamable-http", "url": "https://noolog.io/mcp" }
}
}
That's it — a bot anywhere can search and read the docs. GET /mcp returns a 405 (the channel
is POST-only) — that response itself confirms it's live.
Tools
| Tool | Args | Returns |
|---|---|---|
search_docs |
query (keywords), limit? |
Ranked pages — title, section, url, snippet. |
fetch_doc |
url (from a search hit) |
The full markdown of that page. |
Typical flow: search_docs for the topic → pick a hit → fetch_doc its url → read.
The MCP indexes exactly the pages published here — its answers always match what's in the portal.