MCP server

The endpoint

https://aistats.novusstreamsolutions.com/mcp

One stateless POST, JSON-RPC over HTTP, protocol revision 2026-07-28. No account, no key and no sign-up: every tool reads content this site already publishes, so an authentication step would be theatre. A GET answers 405 on purpose, because that revision removed the server-sent stream and a client opening one needs a definitive answer rather than a request that hangs.

Adding it to a client

Most clients take a block like this. The server name is yours to choose; the URL is the part that matters.

{
  "mcpServers": {
    "novus-ai-stats": {
      "type": "http",
      "url": "https://aistats.novusstreamsolutions.com/mcp"
    }
  }
}

What it exposes

What Novus AI Stats is, and what it does with a filedescribe_product

The product in one call: what it measures, which sources it supports, and where the parsing happens. That last part decides how you can use this server. Exports are read in the visitor's own browser and the transcript never reaches a server, so this endpoint holds no file, no account and no statistic belonging to anyone. It can tell you what the product can measure; it cannot measure anything for you.

What a metric means, and which exports can report itexplain_metric

For one metric or all thirteen: what it measures, exactly how it is calculated, what a high or low reading tells you, how it can mislead, and (per source) whether AI Stats can report it at all, with the honest reason when it cannot. THIS IS THE TOOL FOR "why is this blank?": a metric a format does not carry is reported as unavailable, never as zero, and the reason is a property of the export rather than of the person's usage.

List every site in the Novus networklist_novus_sites

The other Novus Stream Solutions sites and what each one is for. Use this when a request is outside what this site does: file conversion, PDFs, background removal, study material, music video, sample files or food-market data. Each site runs its own MCP server at /mcp.

List the AI tools whose exports this product readslist_supported_sources

Every source AI Stats parses, with the file it takes, its adapter version, whether it is graded Verified or Beta, its known limitations and how many of the thirteen metrics it can report. Also lists the planned adapters, which accept nothing today. Use it to answer "can I import my X?" without guessing.

Read one documentation page or tutorial in fullread_doc

The complete text of one /docs page or one /tutorials chapter, by slug. This is published content (the same bytes /llms-full.txt already serves to any crawler), not anything belonging to a user. Call search_docs first to get a slug.

Find the page, doc, tutorial or article that answers a questionsearch_docs

Search this site in the words a person would use: "why is my token count empty", "how do I export from Claude", "is my transcript uploaded". Ranks the product pages, the documentation, the tutorials, the blog and the per-source integration pages together, and returns the URL of each.

Resources and prompts

Alongside the tools it publishes 2 readable resources and 1 prompt:

Novus AI Stats site index for language modelshttps://aistats.novusstreamsolutions.com/llms.txt

The site index written for language models: every product, integration, doc, tutorial, blog and policy page.

Site index plus the full text of every articlehttps://aistats.novusstreamsolutions.com/llms-full.txt

The index followed by the complete text of every documentation page, tutorial and blog post.

Explain why a number on the dashboard is emptyexplain_a_blank_metric

Walk through answering "why is this metric blank?" using the per-source capability matrix rather than guessing.

What it will not do

It cannot read your transcripts, and that is not a policy this endpoint enforces. It is a fact about where the work happens. An export is parsed in your own browser, in a worker, and the prompts, replies, titles and file contents inside it never reach a server at all. There is therefore no transcript here for a tool to open, no matter who is asking or what they ask for.

It cannot import or analyse a file either. Handing this server a path or a pasted export would not start anything: the parser is browser code, so a tool offering to run it would either be asking you to upload the very thing the product exists not to upload, or reporting a result it never computed.

And it cannot see a signed-in account: not a dashboard, not a project, not a single session. The endpoint is unauthenticated precisely because everything it can reach is already public: which sources are supported and what file each takes, what every metric means and how it is computed, and which metrics a given export cannot report and why. A metric a format does not carry is reported as unavailable, never as zero.