Introducing MCP tools for DAIR.AI. Connect the MCP to your Codex/Claude/Grok Bot to discover and explore the top AI papers. We carefully curate the papers, so you can expect mostly bangers. The index has all the papers I have featured on X over the last couple of years. Get... started: > find and summarize awesome papers > explore our harness engineering collection > find new methods to experiment with > find sota results around AI topics > generate literature review/surveys > build your personal collection > visualize papers > and much more The demo shows how to install it, including some ideas inspired by Andrej Karpathy for visualizing a paper. We are just getting started. Reach out if you have feedback. We are releasing a few benchmarks with this in the next couple of weeks and will ship regular additions and improvements. Stay tuned!show more

@dair_ai does the mcp return the full paper text or just abstracts and metadata. that decides how useful the summarize step actually is

@dair_ai Curious what the MCP actually hands the agent. If it's your summary plus the link, Codex will end up citing your take instead of the paper. Does a tool call return the abstract or full text too?

@dair_ai Citations are key to making this work as we want. We're already doing that, but we will continue to improve it further, as it plays a key role in discovery and a reliable literature review.

@dair_ai Good to know citations are already in. That's what I'd check first in a lit review: does each claim point back to the paper itself, or to the summary.

@dair_ai Thanks for this! I sent a screenshot of your post to Codex and got DAIR setup on my machine with scheduled task updates on AI research + architectures.

@dair_ai That's awesome! Let me know if you see any issues. I'll spend a lot of time optimizing this with benchmarks we are building in-house. I have lots of other papers that I don't get to feature daily, so you might find a few hidden gems.

@dair_ai BTW, I like this automation you put together (architecture stuff). I have something like that for myself, and I am working to make it even better with the MCP tools. More on this soon.

@dair_ai Mostly bangers promised, my reading list is about to get dangerously long

@dair_ai You can reduce that list now by using your agents to organize and take notes. And ideally, implement a few ideas.

@dair_ai Curated papers behind an MCP tool is a neat idea. The thing I'd watch is freshness: once a real account sits behind a tool, stale output that looks current is what breaks first.

@dair_ai this is exactly what I’ve been looking for. “Look at this repo, find a relevant paper, help me test the idea.” finally a way for my agent to give me homework too 😂

@dair_ai It's honestly one of the top ways I use the MCP tools.

can it keep track of new papers on a topic too?

Great question. It does that to some extent. You can see the papers are categorized into high-level categories, but we're rolling out some improvements soon.

will be checking it out

@dair_ai Would use this with a problem already in hand. Find a relevant paper, pick one thing to test, then record what happened. Otherwise my reading list just gets bigger.

@dair_ai Going to try this with my NLP students for their lit reviews. Does it cite the section of the paper it pulls from?

@dair_ai It can do it. It's not perfect, but we are working to improve it in the coming days. If you have any feedback or issues, please DM me or post them in the community forum. BTW, lots of great NLP papers there.

@dair_ai I don't know how you came up with the idea of creating weekly tops, but helps with fomo substantially

@dair_ai does it work with plain keyword search or do you need the exact title? that decides how often the agent finds anything

@dair_ai Should work either way.

@dair_ai this mcp integration makes exploring the archive so much faster

@dair_ai what baffles me is how long we relied on basic search engines for research papers when MCP servers make it this seamless

@dair_ai Any plan to tag which papers held up later? A simple replicated or didn't flag is the part I'd actually use.

@dair_ai curated beats raw arxiv search for this. does the mcp give the agent full text or just abstracts? agents love confidently citing a paper they only read the summary of

@dair_ai Paper MCP is the quiet upgrade. The agent stops guessing what is new and pulls the paper you would have bookmarked at 1am. How do you stop it from citing the flashy one instead of the one that works?

The curation is the actual product here. Raw arXiv search hands the agent 200 papers and a ranking problem; a human-curated index hands it five papers someone actually read. I'd use it the other way too — "find papers that would change how this repo is built" as a standing review.