DeepResearch MCP
ameeralnsLABSCO SUMMARY
The pieces for turning research into a finished document.
Taken together the four cover a whole path — search the web, pull the images and video, then render the result as a PDF or a deck — which is the sequence that usually needs three unrelated tools and a manual export.
What it is
A bundle of four independent MCP servers packaged together for convenience. Each runs as its own process and is registered separately in your client, so you take the ones you want and leave the rest.
What you get
- Media Tools — search images on Unsplash and videos on YouTube, download images, and understand image and video content through Google Gemini and YouTube transcripts
- Information Retrieval — web search through Google Custom Search, and crawling of web pages
- PDF Creator — generate PDF documents from HTML content using Playwright and Pillow
- Presentation Creator — assemble PowerPoint presentations from HTML slide content, and produce PDFs from HTML
Requirements
The three Node servers install from npm as information-retrieval-mcp-server, media-tools-mcp-server and presentation-creator-mcp-server, so npx -y is enough and nothing needs cloning. Keys differ per server: Information Retrieval needs GOOGLE_API_KEY and GOOGLE_CSE_ID from a Custom Search Engine; Media Tools needs UNSPLASH_ACCESS_KEY plus a Google key with the YouTube Data API v3 and the Gemini API enabled; Presentation Creator needs none. The Python servers are built from source in a virtual environment and both require playwright install for their browser, with the client pointed at the interpreter inside that .venv.
Setup effort
One command plus a key — npx -y information-retrieval-mcp-server, then supply credentials
USE CASES
01
Researching a topic with sources
Search the web and crawl the pages that matter, so the assistant is working from the actual text rather than its recollection.
02
Finding images for a piece of work
Search Unsplash and download the ones you picked, without leaving the conversation to browse.
03
Understanding a video without watching it
Search YouTube, pull the transcript, and have the content interpreted rather than summarised from the title.
04
Turning HTML into a shareable PDF
Render the document through the PDF server, which drives a real browser, so what you get matches what the HTML looks like.
05
Building a deck from slide markup
Write the slides as HTML and let the presentation server assemble them into a PowerPoint file.
BEFORE YOU INSTALL
What to know first
Requires
- npx on PATH
- GOOGLE_API_KEY in your environment
- UNSPLASH_ACCESS_KEY in your environment
- GEMINI_API_KEY in your environment
- GOOGLE_CSE_ID in your environment
INSTALLATION
Connect your client
Server setup
TERMINAL
npx -y information-retrieval-mcp-serverMCP Clients
Claude Desktop — npx
CLAUDE DESKTOP — NPX · JSON
{
"mcpServers": {
"information-retrieval": {
"command": "npx",
"args": [
"-y",
"information-retrieval-mcp-server"
],
"env": {
"GOOGLE_API_KEY": "<YOUR_GOOGLE_API_KEY>",
"GOOGLE_CSE_ID": "<YOUR_GOOGLE_CSE_ID>"
}
},
"media-tools": {
"command": "npx",
"args": [
"-y",
"media-tools-mcp-server"
],
"env": {
"UNSPLASH_ACCESS_KEY": "<YOUR_UNSPLASH_ACCESS_KEY>",
"GEMINI_API_KEY": "<YOUR_GEMINI_API_KEY>"
}
},
"presentation-creator": {
"command": "npx",
"args": [
"-y",
"presentation-creator-mcp-server"
]
}
}
}
