Agent Teams PaperSearch
A multi-agent academic paper search framework powered by Claude Code. Five specialized AI agents search 14+ databases in parallel, aggregate results, download PDFs, and find mirror links.
Quick Install
Option A: One-Command Install (Recommended)
claude plugin marketplace add Xinjing-Guo/Agent-Teams-PaperSearch --sparse .claude-plugin plugin
claude plugin install agent-teams-papersearch
Option B: Local Install
git clone https://github.com/Xinjing-Guo/Agent-Teams-PaperSearch.git
claude plugin add ./Agent-Teams-PaperSearch/plugin/agent-teams-papersearch
Then in any Claude Code session:
/paper-search Xinjing Guo All-state model MOSFET
Prerequisites
Required
- Claude Code v2.1.32+ (
claude --version)
- Docker (for paper-search MCP server)
Required MCP Servers
This plugin depends on MCP tools from these servers. All must be configured before use.
1. Paper Search MCP (Required)
Provides 57 search/read/download tools across 22 academic databases.
# Step 1: Install Docker
brew install --cask docker # macOS
# Start Docker Desktop and wait for it to launch
# Step 2: Pull the image
docker pull mcp/paper-search
# Step 3: Verify
docker images mcp/paper-search
If you have the paper-search-tools Claude plugin installed, run:
/paper-search-tools:setup
2. Tavily MCP (Required for ResearchGate search)
Provides web search with domain filtering — used to find ResearchGate mirrors.
# Get a free API key at https://tavily.com
# Then configure:
/tavily-tools:setup
Or manually add to your Claude settings:
{
"mcpServers": {
"tavily": {
"command": "npx",
"args": ["-y", "@anthropic/mcp-tavily"],
"env": {
"TAVILY_API_KEY": "your-key-here"
}
}
}
}
3. Consensus & Scholar Gateway (Optional, cloud)
These are claude.ai built-in MCP integrations. If you use Claude Code with a claude.ai account, they may already be available. Check with /mcp to see if Consensus and Scholar Gateway appear in your server list.
If not available, Scout-Cloud will still work via Tavily/ResearchGate search.
Verify Setup
After configuring, restart Claude Code and run:
/mcp
You should see these servers connected:
paper-search (or plugin:paper-search-tools:paper-search)tavily (or plugin:tavily-tools:tavily)Consensus (optional)Scholar Gateway (optional)
Team Members
| Agent | Role | Databases | Color |
|---|
| Marshall | Coordinator | — | — |
| Scout-Core | Tier 1 Search | Google Scholar, Semantic Scholar, OpenAlex, CrossRef | Green |
| Scout-Domain | Tier 2 Search | arXiv, PubMed, bioRxiv, medRxiv, DBLP, SSRN, IACR, OpenAIRE, BASE, CORE, DOAJ, Zenodo, Unpaywall, PMC, EuropePMC | Cyan |
| Scout-Cloud | Tier 3 Search | Consensus, Scholar Gateway, ResearchGate (via Tavily) | Magenta |
| Librarian | Download | curl (arXiv, bioRxiv, direct PDF), download_with_fallback | Yellow |
| Archivist | Aggregation | Deduplication, ranking, citation verification | Blue |
Workflow
Phase 1: Parallel Search → 3 scout agents search 14+ databases simultaneously
Phase 2: Aggregation → Deduplicate by DOI/title, rank by citations
Phase 3: Presentation → Formatted tables with database coverage
Phase 4: Download (on ask) → curl download or manual links provided
Features
- 14+ academic databases searched in parallel
- ResearchGate mirror detection via Tavily web search
- Smart topic routing — domain databases selected by keyword analysis
- Citation verification — never claims unverified citation relationships
- Download with fallback — curl > MCP > manual links (always provides something)
- Docker-aware — knows MCP download tools lose files, uses curl instead
Known Limitations
