Lune

Lune Research is a tool from Retrograde Labs that works as an MCP server for research. It sits between an AI assistant and the computer science literature. You connect it once, then ask your agent a question in plain English. The agent calls Lune's tools and gets paper records back: title, authors, venue, year, abstract, and a link that opens. When Lune has permission to index a paper's full text, the agent can read the relevant sections and quote them.

The point is to stop a model from answering from memory. Ask a chatbot for the foundational papers on retrieval-augmented generation and it may hand you a confident citation that doesn't exist. Lune forces a retrieval step first, and every claim in the answer traces back to a paper record you can inspect.

Coverage is the trade-off. Lune indexes conferences ranked A or A* by CORE and CCF, from 2016 onward. AI and machine learning venues such as NeurIPS, ICML, ICLR, ACL and CVPR are in. So are security conferences like CCS and USENIX Security, systems work from OSDI and SOSP, data papers from KDD, SIGMOD and VLDB, plus WWW and STOC. Workshop papers, tutorials and short papers stay out. That focus helps when you want accepted literature and nothing else, and frustrates when the paper you need sits outside those venues.

Getting Started

1. Sign in at luneresearch.com with Google or an email address. New accounts come with 300 welcome credits.
2. Open the Install page in the dashboard and pick your app. The list covers Claude, Claude Code, ChatGPT, Codex, Cursor, VS Code, GitHub Copilot and other MCP clients.
3. Remote apps connect to https://mcp.luneresearch.com and sign in through OAuth. Local clients run npx -y @retrograde-labs/lune-mcp-server with an access key in LUNE_API_KEY.
4. Ask a question in the app. The agent picks the tool itself, usually search_papers first, then get_paper_fulltext or a citation lookup.
5. Check the dashboard when a limit appears. It shows the current plan, today's remaining allowance and the credit balance.