add mcp-for-research blog post #3021

dylanebert · 2025-08-11T16:58:19Z

Added blog post for Research Tracker MCP philosophy, how-to, and pointing toward more resources

ariG23498

LGTM!

mcp-for-research.md

dylanebert · 2025-08-12T15:00:37Z

LGTM!

thanks for the review!

merveenoyan

lgtm 💗💗

merveenoyan · 2025-08-13T06:13:50Z

mcp-for-research.md

+
+Academic research involves frequent **research discovery**: finding papers, code, related models and datasets. This typically means switching between platforms like [arXiv](https://arxiv.org/), [GitHub](https://github.com/), and [Hugging Face](https://huggingface.co/), manually piecing together connections.
+
+The [Model Context Protocol (MCP)](https://huggingface.co/learn/mcp-course/unit0/introduction) is a standard that allows agentic models to communicate with external tools and data sources. For research discovery, this means AI can use research tools through natural language requests, automating platform switching and cross-referencing.


we could add a bit of research tracker MCP Space here, it's too abstract otherwise

Yes, I agree! We could even show a gif with an example from Cursor.

merveenoyan · 2025-08-13T06:14:46Z

mcp-for-research.md

+<gradio-app theme_mode="light" space="dylanebert/research-tracker-mcp"></gradio-app>
+
+## Learn More
+


one sentence call for action would be great

pcuenca

Looking good! I'd suggest to provide some continuity among the sections: why we are doing this, what's in it for the reader, how they can use or extend the tools, etc.

pcuenca · 2025-08-13T14:18:08Z

_blog.yml

+  title: "MCP for Research: How to Connect AI to Research Tools"
+  author: dylanebert
+  thumbnail: /blog/assets/mcp-for-research/thumbnail.png
+  date: Aug 11, 2025


Reminder to update before release

pcuenca · 2025-08-13T14:20:36Z

mcp-for-research.md

+
+At the lowest level of abstraction, researchers search manually and cross-reference by hand.
+
+```bash


Maybe enclose in html <xmp> / </xmp> tags to avoid weird syntax highlighting. Or use > to blockquote.

pcuenca · 2025-08-13T14:22:25Z

mcp-for-research.md

+results = gather_research_info("https://arxiv.org/abs/2103.00020")
+```
+
+The [research tracker](https://huggingface.co/spaces/dylanebert/research-tracker) demonstrates systematic research discovery built from these types of scripts.


Perhaps you could briefly explain somewhere that you've been using this approach to keep track of research in the 3D space, just to make it more relatable.

pcuenca · 2025-08-13T14:27:24Z

mcp-for-research.md

+- Include performance benchmarks when available
+```
+
+The AI orchestrates multiple tools, fills information gaps, and reasons about results:


Suggested change

The AI orchestrates multiple tools, fills information gaps, and reasons about results:

Because the LLM knows the details of what each tool can do, it's able to orchestrate calls to multiple tools, fill information gaps, reason about results and collate them for the user:

pcuenca · 2025-08-13T14:27:59Z

mcp-for-research.md

+
+user: "Find papers related to vision transformers with available code and models"
+ai: # Combines multiple tools to gather complete information
+```


It could be cool to show a summary of an actual conversation, or a link to it, or a video.

pcuenca · 2025-08-13T14:31:20Z

mcp-for-research.md

+- Quality depends on the implementation
+- Understanding the lower layers (both manual and scripted) leads to better implementations
+
+## Setup and Usage


Up to this point it's been a more or less theoretical discussion. Perhaps we could segue into this by reminding the reader that we used a set of custom scripts that evolved into a set of MCP tools that anyone can use, build upon or use for inspiration.

pcuenca · 2025-08-13T14:32:54Z

mcp-for-research.md

+
+### Quick Setup
+
+For setup instructions, see the [Research Tracker MCP](https://huggingface.co/spaces/dylanebert/research-tracker-mcp) space, then click `Use via MCP or API` at the bottom of the page. This will provide instructions for adding the server with SSE, i.e.:


Suggested change

For setup instructions, see the [Research Tracker MCP](https://huggingface.co/spaces/dylanebert/research-tracker-mcp) space, then click `Use via MCP or API` at the bottom of the page. This will provide instructions for adding the server with SSE, i.e.:

For setup instructions, see the [Research Tracker MCP](https://huggingface.co/spaces/dylanebert/research-tracker-mcp) space, then click `Use via MCP or API` at the bottom of the page. This will provide instructions for adding the tools as a server you can integrate with your preferred IDE, such as Cursor or Claude Code:

pcuenca · 2025-08-13T14:33:21Z

mcp-for-research.md

+<gradio-app theme_mode="light" space="dylanebert/research-tracker-mcp"></gradio-app>
+
+## Learn More
+


pcuenca · 2025-08-13T14:34:32Z

mcp-for-research.md

+
+Academic research involves frequent **research discovery**: finding papers, code, related models and datasets. This typically means switching between platforms like [arXiv](https://arxiv.org/), [GitHub](https://github.com/), and [Hugging Face](https://huggingface.co/), manually piecing together connections.
+
+The [Model Context Protocol (MCP)](https://huggingface.co/learn/mcp-course/unit0/introduction) is a standard that allows agentic models to communicate with external tools and data sources. For research discovery, this means AI can use research tools through natural language requests, automating platform switching and cross-referencing.


Yes, I agree! We could even show a gif with an example from Cursor.

pcuenca · 2025-08-13T14:36:57Z

mcp-for-research.md

+
+The [Model Context Protocol (MCP)](https://huggingface.co/learn/mcp-course/unit0/introduction) is a standard that allows agentic models to communicate with external tools and data sources. For research discovery, this means AI can use research tools through natural language requests, automating platform switching and cross-referencing.
+
+## Research Discovery: Three Layers of Abstraction


Suggested change

## Research Discovery: Three Layers of Abstraction

This post shows how we built a set of MCP tools for research, how you can use them for your own research projects, or as inspiration for new tools adapted to your needs.

## Research Discovery: Three Layers of Abstraction

julien-c

Could be interesting to have a quick review from @evalstate

evalstate

Great article! would love to work on getting this working well with the other hf mcp tools; @Vaibhavs10 - looks like might be a good addition to mcp-with-openai agents.

evalstate · 2025-08-14T08:40:33Z

mcp-for-research.md

+{
+  "mcpServers": {
+    "gradio": {
+      "url": "https://dylanebert-research-tracker-mcp.hf.space/gradio_api/mcp/sse"


https://dylanebert-research-tracker-mcp.hf.space/gradio_api/mcp/
submitted a PR to the space to upgrade gradio version and make Streamable HTTP the default (which would change this URL).

evalstate · 2025-08-14T08:51:49Z

mcp-for-research.md

+
+### Quick Setup
+
+For setup instructions, see the [Research Tracker MCP](https://huggingface.co/spaces/dylanebert/research-tracker-mcp) space, then click `Use via MCP or API` at the bottom of the page. This will provide instructions for adding the server with SSE, i.e.:


I'd also recommend mention adding it to the Hugging Face MCP Server via https://huggingface.co/settings/mcp (i've just tested this and it works fine) - that will give you access to complementary model, dataset details and search too via one endpoint (we also have directory listings for Claude, VSCode, Cursor with one-click installs).

i'm also working on https://github.com/huggingface/blog/blob/feat/claude-images-mcp/claudeimages-mcp.md which will show how to set this up in claude.

Vaibhavs10

Brilliant work @dylanebert - just want to emphasis what @evalstate also mentioned that let's use hf.co/mcp i.e. the HF MCP server as the standard way for people to use spaces as tools in claude/ cursor/ vscode (wherever they want to use)

Co-authored-by: Aritra Roy Gosthipaty <[email protected]>

dylanebert · 2025-08-14T17:07:30Z

Thanks a lot for the reviews! I've set the publish date to next Monday in case there's more integrations to add/test.

Revisions:

@pcuenca @merveenoyan I added a GIF at the top for immediate context, and added transition paragraphs for the abstraction sections, and a call-to-action at the end

@evalstate @Vaibhavs10 Replaced instructions to hf-mcp-first instructions

dylanebert requested a review from pcuenca August 11, 2025 16:58

ariG23498 approved these changes Aug 12, 2025

View reviewed changes

mcp-for-research.md Outdated Show resolved Hide resolved

mcp-for-research.md Outdated Show resolved Hide resolved

merveenoyan approved these changes Aug 13, 2025

View reviewed changes

pcuenca approved these changes Aug 13, 2025

View reviewed changes

julien-c reviewed Aug 14, 2025

View reviewed changes

evalstate reviewed Aug 14, 2025

View reviewed changes

Vaibhavs10 reviewed Aug 14, 2025

View reviewed changes

dylanebert and others added 4 commits August 14, 2025 13:04

add mcp-for-research blog post

f7b5bae

Update mcp-for-research.md

dc05c5d

Co-authored-by: Aritra Roy Gosthipaty <[email protected]>

move embedded space down

9f59785

incorporated suggestions

a6d6cc7

dylanebert force-pushed the main branch from b151a7b to a6d6cc7 Compare August 14, 2025 17:04


		Academic research involves frequent research discovery: finding papers, code, related models and datasets. This typically means switching between platforms like [arXiv](https://arxiv.org/), [GitHub](https://github.com/), and [Hugging Face](https://huggingface.co/), manually piecing together connections.

		The [Model Context Protocol (MCP)](https://huggingface.co/learn/mcp-course/unit0/introduction) is a standard that allows agentic models to communicate with external tools and data sources. For research discovery, this means AI can use research tools through natural language requests, automating platform switching and cross-referencing.

		<gradio-app theme_mode="light" space="dylanebert/research-tracker-mcp"></gradio-app>

		## Learn More


		At the lowest level of abstraction, researchers search manually and cross-reference by hand.

		```bash

	The AI orchestrates multiple tools, fills information gaps, and reasons about results:
	Because the LLM knows the details of what each tool can do, it's able to orchestrate calls to multiple tools, fill information gaps, reason about results and collate them for the user:


		### Quick Setup

		For setup instructions, see the [Research Tracker MCP](https://huggingface.co/spaces/dylanebert/research-tracker-mcp) space, then click `Use via MCP or API` at the bottom of the page. This will provide instructions for adding the server with SSE, i.e.:


		The [Model Context Protocol (MCP)](https://huggingface.co/learn/mcp-course/unit0/introduction) is a standard that allows agentic models to communicate with external tools and data sources. For research discovery, this means AI can use research tools through natural language requests, automating platform switching and cross-referencing.

		## Research Discovery: Three Layers of Abstraction

add mcp-for-research blog post #3021

Are you sure you want to change the base?

add mcp-for-research blog post #3021

Uh oh!

Conversation

dylanebert commented Aug 11, 2025

Uh oh!

ariG23498 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

dylanebert commented Aug 12, 2025

Uh oh!

merveenoyan left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pcuenca left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

julien-c left a comment

Choose a reason for hiding this comment

Uh oh!

evalstate left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Vaibhavs10 left a comment

Choose a reason for hiding this comment

Uh oh!

dylanebert commented Aug 14, 2025

Uh oh!

Uh oh!