Paper Lantern
Research intelligence for AI coding agents. Surfaces what 2M+ CS papers say at any technical decision your agent hits.
What it does
Your agent's training data freezes at a point in time. Paper Lantern gives it on-demand access to 2M+ CS research papers with evidence, tradeoffs, and implementation guidance. When your agent is about to pick a chunking strategy, a reranker, an eval method, or an agent memory pattern, it asks Paper Lantern what recent research says.
What it's used for
Prompting, evals, RAG (chunking / retrieval / reranking), agent memory, routing, fine-tuning, test generation, extraction, classification, code review, algorithm choice, architecture trade-offs. Any technical decision where recent CS research improves the outcome.
Tools
| Tool | What it does |
|---|---|
explore_approaches | Survey 4-6 approach families with evidence and tradeoffs |
deep_dive | Investigate one technique in depth (implementation, hyperparameters, failure modes) |
compare_approaches | Side-by-side comparison of 2-3 candidates |
check_feasibility | GO / PROTOTYPE / RECONSIDER verdict given constraints |
give_feedback | Tell us what worked |
When it activates vs when it doesn't
- Activates: technique and architecture decisions where research evidence improves the outcome
- Skips: syntax questions, library API lookups, debugging, code formatting, general programming tasks
Install
npx paperlantern@latest
OAuths via device code, mints a pl_* API key, writes MCP config for Claude Code, Cursor, Windsurf, GitHub Copilot (VS Code), Codex, and Gemini CLI.
Manual server config
{
"mcpServers": {
"paper-lantern": {
"type": "http",
"url": "https://mcp.paperlantern.ai/chat/mcp?key=YOUR_KEY"
}
}
}
Links
- Website: paperlantern.ai/code
- Docs: paperlantern.ai/docs
- Official MCP Registry:
ai.paperlantern/code - Source: github.com/paperlantern-ai/paperlantern-cli
## Also published on
- Official MCP Registry: `ai.paperlantern/code` (published 2026-04-21)
- PulseMCP: auto-ingests from the official registry (appears within ~1 week)
Server Config
{
"mcpServers": {
"paper-lantern": {
"type": "http",
"url": "https://mcp.paperlantern.ai/chat/mcp?key=YOUR_KEY"
}
}
}