Notes from the deeplearning.ai course on box integration with A2A and MCP server

Course link - https://learn.deeplearning.ai/courses/build-ai-apps-with-mcp-server-working-with-box-files/lesson/dew9nq/introduction-to-box-mcp-server

Toy solution using local databse

in the code base for L1, we are doing the following -

  1. copy the invoices from a folder /invoice
  2. parse the pdfs invoices using pdf parser PdfReader from PyPDF
  3. for every file, read the pdf, and pass to gemini and extract the following in json format -
    1. client name
    2. invoice amount
    3. product name
  4. Then these json are read and inserted into a db placed in the local.
  5. Then aggregation is done from the db.
  6. code - https://github.com/samratkar/samratkar.github.io/blob/main/_posts/concepts/genai/notes-codes/a2a-box/l1/L1.ipynb

Scalable solution using invoices from Box using MCP server

  1. Box MCP server takes care of the following
    1. list folder contents
    2. text / data extraction
    3. takes care of OCR and text as well internally.
    4. fast : AI agents can directly access, understand, and act on content in the box. no need to download.
    5. the box mcp server is scalable and handle large number of files.
    6. built in search and extraction is supported in the box mcp server. no need to worry about OCR, or prompting the model for specific fields. they are available in prebuilt functions.
  2. MCP Server box code github