Knowra

Paper search and Scout briefs for your AI agent. Methods, evidence and scope to help decide what deserves a closer read.

[Knowra @knowra_app]

Knowra MCP now supports OAuth.

Paste one URL into Claude or ChatGPT, sign in, and your assistant can search arXiv and read Scout reports. No API keys, no config files.

https://t.co/dBDtvP8Iab

[RexZhang @Reeexopia]

dsh-plugin-knowra is out. Search arXiv by topic, author or exact dates from DeepSeek Harness, then let Scout read the full paper — method, evidence, limitations — so your agent can tell you which papers are worth a close read.

https://t.co/GlKfj5H7C3

[Knowra @knowra_app]

@ktwu01 看了 Today，给每条进展写清楚“对评测决策有什么影响”这个角度很实用。你自己做 Agent Benchmark 时，最想提前发现的是任务/能力重叠，还是数据集和评测协议已经有人做过？感觉这会直接影响追踪时该读到哪一层。

[Knowra @knowra_app]

@umarsaif How do you pick the six for a weekend: follow one research question across papers, or choose a broad shortlist? Those seem like quite different reading tasks. Curious which sources help you decide what earns a full read.

[RexZhang @Reeexopia]

Self-improvement work asks what to train on next. A-RSI asks when current tasks stop teaching.

It refreshes the task pool when fewer prompts yield nonzero group advantage. No val scores or human labels: 57.77 → 61.05 on 11 benchmarks.

Picking by difficulty didn't beat random.

[RexZhang @Reeexopia]

In agentic RL, credit usually goes per token (PPO) or per rollout (GRPO). A PRM would give step-level credit, but agent turns are hard to score.

MPO builds an agent-shaped "PRM" from one critic: GAE across turns, a mean-preserving split within. Beats GRPO/GSPO/PPO at 4B/30B-A3B.

[Knowra @knowra_app]

@DivyanshT91162 Good list. Missing piece for me: "what's new on arXiv in my area this week, and which of it holds up?"

We built Knowra for that. Your agent searches arXiv by exact date, then reads each paper's methods and limits before recommending. (Hosted MCP, not a repo.)

[Knowra @knowra_app]

Nature reports arXiv got 40,363 submissions in September, about double two years ago.

The cap helps moderators. Readers still face the firehose.

Our fix: filter to the exact days, then check what each paper actually tested. That's why we built Knowra's Scout.

[arXiv.org @arxiv]

arXiv has updated our policy on rate limiting for all submitters.

This update was made to fairly distribute moderator time & support the arXiv community of staff, volunteers, readers & authors.

Please read our announcement to learn more: https://t.co/lvzWuiv0hx

[Knowra @knowra_app]

Knowra is now in Claude's connector directory.

Find it in the directory and click Add. No URL to paste.

Search arXiv for free, and sign in when you want Scout reports on a paper's methods, evidence and limits.

https://t.co/OATPl7Z7vK

[Knowra @knowra_app]

@curiouswavefn Fair point. An arXiv connector should help agents filter papers, not just collect citations.

So we built Knowra: selective search + Scout to examine claims, evidence, and limitations.

Not a substitute for peer review, of course. I'm one of the devs: https://t.co/eKXvgwfQTJ

[Knowra @knowra_app]

@sidgairo18 If you want arXiv specifically, we built Knowra for this. Paste https://t.co/dBDtvP8Iab into Claude or ChatGPT as a custom connector and sign in. Search is free. (I’m the dev, happy to hear what breaks.)

[Knowra @knowra_app]

Scout reads the full paper and flags claims, evidence and limits, so you can tell which papers are worth a close read.

Add it once on web or desktop and it works in the mobile apps too.
Search is free; free accounts get 5 Scout reports a day.

https://t.co/eKXvgwfQTJ

[Knowra @knowra_app]

Knowra MCP now supports OAuth.

Paste one URL into Claude or ChatGPT, sign in, and your assistant can search arXiv and read Scout reports. No API keys, no config files.

https://t.co/dBDtvP8Iab

[Knowra @knowra_app]

Claude: Settings → Connectors → Add → Add custom connector → paste the URL → Connect

ChatGPT: Customize → Plugins → Add → Add custom MCP server → paste the URL (OAuth) → Create → Connect

[Knowra @knowra_app]

New to MCP? Save image 2 and give it to your agent.

Ask it to help you connect Knowra, then try a research question: search papers, shortlist from abstracts, and read a relevant Scout.

Start here: https://t.co/fvIKD6mFck

[Knowra @knowra_app]

3/3

Paper search needs no account. Free accounts can unlock 5 Scouts per UTC day. Rereading a paper you've unlocked uses no more quota.

Create a key, copy the MCP config to your agent, and check the connection:
https://t.co/fvIKD6mFck

[Knowra @knowra_app]

1/3

Knowra MCP is live.

Search papers in your AI agent, then use Scout briefs to decide what deserves a closer read.

Scout draws on paper content to surface methods, evidence and limitations.

Try it: https://t.co/eKXvgwfQTJ

[Knowra @knowra_app]

2/3

Scout gives your agent methods, evidence, scope and checks to follow up in the paper.

The image shows Scout in the Knowra App. Over MCP, your agent receives text, figure captions and descriptions.

Paper links let you check the source.