early experiment · read-only · source-grounded

ask my public notebook.

this can search what i have written and cite the pages. it cannot reliably think like me yet. treat every answer as a documented belief, an inference, or missing context.

the notebook connection is liveit reads only this public website. it cannot see private data, make decisions for me, or write or change anything.
https://angaramoksh.com/mcp-server

paste this into chatgpt, claude, grok, cursor, codex, or another client that accepts remote mcp servers.

four steps. that is it.

  1. copy the endpoint above.
  2. open your ai client’s connectors, integrations, or mcp settings.
  3. add a remote mcp server and paste the url. no authentication is required.
  4. ask a real question. tell it to cite the pages it used.

some products hide custom mcp connections behind particular plans or workspace settings. if one client does not expose the option, use another remote-mcp-compatible client.

what this should feel like

i have a usable product but almost no customers. what has he actually written?
documented belief: he argues that a good product sitting inside a laptop still makes zero rupees. nail distribution instead of polishing forever.

inference: pick one narrow audience and measure conversations before adding another feature. the notes do not name your market.

source: distribution over product

steal one of these questions

what your ai can actually read

get_site_infoBio, ventures, topics, and public contact links
list_notesAll notes with titles, sections, and summaries
get_noteFull text of one note
list_pagesAbout, Work, People, Gallery, and other public pages
get_pageFull content of one public page
searchRanked full-text search with source URLs

or try ordinary ai search now

these buttons only open a normal search with the question filled in. the mcp endpoint above is the live, complete route.