# Paperclip
Paperclip is an agent-native CLI for searching, retrieving, and synthesizing insights from 8M+ biomedical papers. Built by Generative Expert Labs (gxl.ai), it targets the gap left by stateless literature APIs; agents need persistent context to deep-dive across a corpus, not one-shot searches.
## Commands
- `search` — hybrid BM25 + embedding retrieval with TL;DR summaries
- `grep` — corpus-wide regex matching, reportedly 36-294x faster than native grep
- `map` — parallel query across many papers (~8x speedup vs sequential)
- `ask-image` — VLM analysis of paper figures without downloading
- `sql` — read-only metadata queries (author, journal, date, source)
- `from` — stateful persistence of result sets for chained operations
## Why it matters
Most agentic research tools treat each API call as isolated. Paperclip's `from` command persists paper subsets between calls, so agents can narrow a search, then grep, then map across the same set without rebuilding context each time. This turns literature review into a composable pipeline.
## Integration
- MCP server; integrates directly with [[Claude Code]] and other MCP clients
- Complements Sy, gxl.ai's human-facing literature search agent
## References
- Blog post: https://gxl.ai/blog/paperclip
- Website: https://gxl.ai
## Related
- [[Model Context Protocol (MCP)]]
- [[Claude Code]]
- [[Large Language Models (LLMs)]]
- [[Embeddings]]