Connect an AI assistant
Claude, ChatGPT and any MCP client can search this site, see what we charge, estimate a project and send us an enquiry. This page says what moves where, and the limits we set.
The tools
Five tools, four of them read-only
Every assistant sees the same five names and descriptions, each marked read-only or write, so you know what a tool can do before it runs.
Search National Digital pages
Finds published pages for a topic or keyword: services, guides, case studies and pricing. Read-only.
Read a National Digital page
Returns one published page as clean text, with its title, address and last-modified date. Read-only.
Get services and pricing
What we build and support, and the published price range for each kind of project, in AUD. Read-only.
Estimate a project cost range
An indicative range for a project type and scale, from the same calculation as the estimator on this site. Read-only.
Send an enquiry to National Digital
Sends your name, email and message to the team, once you have asked the assistant to make contact. The only tool that writes.
Claude, ChatGPT and other assistants can read this site and contact us for you. This page covers what they can do, what data moves, and what we hold them to. If you are the person in IT deciding whether to approve the connector, it was written for you.
Try it
If your assistant can connect to MCP servers, paste this:
Connect to the MCP server at https://mcp.nationaldigital.com.au/mcp (Streamable HTTP, no login). Then tell me what National Digital would charge to build a customer portal for a 60-person wholesaler, and draft an enquiry for me to approve before you send it.Claude Desktop, Claude Code, ChatGPT in developer mode and Cursor all take a server URL this way. If yours needs the server added in settings first, the endpoint is the same URL.
What the connector is
We run a remote MCP server. MCP, the Model Context Protocol, is the open standard assistants use to call tools on another service. Ours sits in front of the published pages on this site. There is no account and no login, because everything it can read is already public.
The endpoint is:
https://mcp.nationaldigital.com.au/mcpIt is listed in the official MCP Registry as au.com.nationaldigital/website and submitted to the Claude connector directory under the name National Digital.
How to connect
- Claude: Settings, Connectors, Add custom connector, then paste the endpoint. Once the directory listing is live you can add it from there.
- ChatGPT: add it as a custom connector in developer mode with the same URL.
- Any other MCP client: add a Streamable HTTP server at that URL with no authentication.
Nothing installs on your machine. The assistant calls the server over HTTPS when it needs something.
What data moves, and where
The four read tools send the server a search query, a page identifier, or answers to the estimator questions. The server answers from a snapshot of the published site: the same pages, price bands and estimator rules a visitor sees in a browser. It never follows a link it is handed and cannot reach a page we have not published.
The server keeps no account, sets no cookie and builds no profile. Like any web server it sees the connecting address, which for a hosted assistant is the assistant's own. The enquiry gate counts enquiries per address per day and keeps a one-way hash of each enquiry so a repeat is not sent twice. Both reset daily.
An enquiry is the one thing that leaves the server. It carries a name, an email address, an optional organisation and the message, and reaches our inbox through the same form service as the contact form. The privacy policy names that service and what it holds:
The rules around enquiries
The enquiry tool is the only one that writes. The assistant is told to use it only when you have asked it to make contact, and Claude asks you to confirm before any tool that writes.
- One enquiry per person. A repeat of the same message is acknowledged and not sent again.
- The channel takes a small number of enquiries a day in total, and a share of that per connecting address. When the day's budget is spent the tool says so and points to the contact form. It does not queue the message.
- Messages with several links, or a link where a name should be, are refused as spam.
- If the channel is abused we close it without touching the read tools, and the assistant is told the enquiry was not received.
A person reads every enquiry and replies within one business day, Australian eastern time. The usual routes reach the same team:
Contact form
enquiries@nationaldigital.com.au
Reading this site as text
Every page has a Markdown copy. Ask for any page with the header Accept: text/markdown and the same URL returns the copy: the words on the page, with the navigation and markup left out, and a Content-Location header naming the file.
curl -H "Accept: text/markdown" https://nationaldigital.com.au/services/The copies also sit under /okf/ in the Open Knowledge Format, and each page points to its own with a <link rel="alternate" type="text/markdown"> tag.
llms.txt is the short index: who we are, what an agent can do here, indicative project ranges and a link to every page. llms-full.txt adds each page's summary, for a reader that wants one file.
For developers and agents
The server speaks the current MCP Streamable HTTP transport and keeps no session state, so any client that supports remote servers can use it. Every published page is also exposed as an MCP resource, and the server card is at the standard well-known location on the same host.
The in-browser WebMCP tools on every page of this site, and an HTTP path for an agent that cannot run a browser, are documented in llms.txt, linked above.
Discovery files
- /.well-known/ard.json: a manifest in the Agentic Resource Discovery format, listing what is on this page and the questions each item answers.
- /feed.xml: an Atom feed of the latest published pages.
- /sitemap.xml: every indexable URL and when it last changed.
- /robots.txt: AI crawlers are allowed, and the file says so.
- /.well-known/security.txt: how to report a security problem.