Knowledge MCP

About Knowledge MCP

Knowledge MCP is our internal search for operations runbooks and product specifications. Ask a question in plain English and it shows the exact section of the right document, with the document, page and heading it came from, so every answer can be checked against the source.

What it covers

• The PM Confluence space: production monitoring, alerts, batch jobs, funding, payments and how-to pages
• The NLS Web Service API and NLS XML Import specifications
• Production process and funding runbooks, including their SQL queries
• Any file you add in step 2: PDF, Word, Excel, text, Markdown, source code, or a .zip of them

How it works

1. Add Upload a file, or paste a Confluence link. The PM space is also re-read every 30 minutes.
2. Read PDFs are read with Azure AI Content Understanding. Headings, tables and code are kept, and SQL stays exactly as written.
3. Split Each document is cut into sections that keep their heading path and page number.
4. Store Each section is saved in Azure Cosmos DB with a note of what it means, so it can be found by meaning as well as by its words.
5. Search Search by meaning and by exact words runs together, and Cohere Rerank puts the best match first.
6. Answer The matching section is shown with where it came from. A weak match is refused instead of guessed.

Where to use it

• On this page: verify your work email in step 1, add documents in step 2, ask in step 3.
• In Claude Code, through the MCP tools: search the documents, ask with citations, or list what can be searched.

Good to know

• It shows the passage from the document; it does not write an answer of its own.
• Text inside pictures and diagrams in a PDF is not searchable yet.
• Everyone who can open this page can search everything that has been added.
• It is available on the company network (VDI).