Key Takeaways
- 5 skill libraries provide AI-assisted literature search for coding agents
- K-Dense literature skills: PubMed, Semantic Scholar, bioRxiv integration with citation network analysis
- Orchestra autoresearch: Full-loop literature review as part of the research pipeline
- LabClaw reference skills: Citation formatting, reference management, bibliography generation
- Custom MCP servers: Connect to any database via Model Context Protocol
- Best approach: combine domain-specific skills (K-Dense) with general research skills (Orchestra)
Why AI Literature Search?
Literature search is one of the most time-consuming parts of research. AI agents can automate query construction, result filtering, citation network exploration, and summary generation — turning hours of manual search into minutes.
Core challenge: Different databases have different APIs, query syntax, and coverage. No single skill covers all sources.
Comparison Table
| Skill Library | Databases | Strengths | Limitations |
|---|---|---|---|
| K-Dense Literature | PubMed, Semantic Scholar, bioRxiv | Citation network analysis, domain expertise | Biology-focused |
| Orchestra Autoresearch | Semantic Scholar, arXiv | Full research loop integration | ML/AI-focused |
| LabClaw Reference | CrossRef, DOI.org | Citation formatting, bibliography | No search, formatting only |
| Custom MCP | Any (configurable) | Maximum flexibility | Requires setup |
| Built-in Agent | Web search | Zero setup | Shallow, no API access |
K-Dense Literature Skills
Best for biomedical researchers. Skills include:
- PubMed search: MeSH term optimization, date filtering, study type filtering
- Semantic Scholar: Citation network traversal, influential citation identification
- bioRxiv/medRxiv: Preprint search with version tracking
- Citation analysis: Find seminal papers, identify research gaps, map citation networks
Orchestra Autoresearch Literature Module
Best for ML/AI researchers. Part of the autoresearch meta-skill:
- Searches Semantic Scholar and arXiv
- Generates structured literature summaries
- Identifies gaps in existing literature
- Feeds directly into experiment design and paper writing stages
Building Custom Literature MCP Servers
For databases not covered by existing skills:
// Example: Custom PubMed MCP server
const server = new MCPServer({
tools: [{
name: "pubmed_search",
description: "Search PubMed with MeSH terms",
parameters: { query: "string", maxResults: "number" },
handler: async ({ query, maxResults }) => {
// NCBI E-utilities API call
}
}]
});FAQ
Q1: Which library should I start with?
Biomedical researchers: K-Dense literature skills. ML researchers: Orchestra autoresearch. General: start with built-in web search, then add specific skills as needed.
Q2: Can I combine multiple skill libraries?
Yes. Install K-Dense for domain-specific search and Orchestra for research workflow. They don't conflict.
Q3: How do AI literature searches compare to manual PubMed searches?
AI agents construct better queries (MeSH term expansion, synonym handling), explore citation networks automatically, and generate structured summaries. Manual search is still better for nuanced conceptual queries.
Summary
No single skill library covers all literature search needs. The best approach combines domain-specific skills (K-Dense for biomedicine, Orchestra for ML) with custom MCP servers for specialized databases. AI-assisted literature search excels at query optimization, citation network exploration, and structured summarization — complementing rather than replacing manual expert search.