Scholar Agent

中文

General-purpose LLMs are often inaccurate and outdated in specialized domains. Scholar Agent combines online research + local knowledge accumulation into a sustainable knowledge flywheel, making your AI smarter in your domain over time. It also builds a human-readable knowledge base for quick learning. Integrates seamlessly with Claude Code and VS Code Copilot via MCP.

What It Does

Your question
    │
    ▼
Online research (LLM web search + academic APIs)
    │
    ▼
Structured synthesis (with citations, confidence, uncertainty)
    │
    ▼
Local accumulation (Markdown knowledge cards + BM25 index)
    │
    ▼
Next question: AI checks local first ── hit? ──► use directly, fast & accurate
    │ miss
    ▼
Research again → accumulate → reindex ──► knowledge base keeps growing

Each round compounds. Knowledge cards have full lifecycle management: draft → reviewed → trusted → stale → deprecated.

Academic Research Pipeline

Scholar Agent includes a comprehensive academic paper research pipeline:

• Paper Search — Search papers from arXiv, DBLP, and Semantic Scholar. Filter by top conferences (CVPR, ICCV, ECCV, ICLR, AAAI, NeurIPS, ICML, ACL, EMNLP, MICCAI)

• Smart Scoring — Four-dimensional scoring engine (relevance, recency, popularity, quality) ranks papers by your research interests

• Deep Analysis Notes — Auto-generate 20+ section Obsidian-style markdown notes with <!-- LLM: --> placeholders for AI-assisted completion

• Figure Extraction — Extract images from arXiv source archives and PDFs (via PyMuPDF)

• Daily Recommendations — Automated daily paper search, scoring, deduplication, and recommendation note generation

• Paper → Knowledge Card — Convert paper analyses into knowledge cards that feed back into the knowledge flywheel

• Keyword Auto-Linking — Scan notes for technical terms and create [[wiki-links]] automatically

Quick Start

Embed into an existing project

cd my-project && git clone https://github.com/zfy465914233/scholar-agent.git
bash scholar-agent/setup.sh
# Restart Claude Code to activate

This will create the directory structure, copy config templates, install skills, and build the knowledge index.

Use as a standalone project

# Clone and install
git clone https://github.com/zfy465914233/scholar-agent.git
cd scholar-agent
pip install -r requirements.txt

# Build the knowledge index
python scripts/local_index.py --output indexes/local/index.json

MCP configs are pre-configured:

• Claude Code: .mcp.json is ready. cd into the project and start Claude Code.

• VS Code Copilot: .vscode/mcp.json is ready. Open the project, enable agent mode.

MCP Tools

Core Tools (always available)

Academic Tools (set SCHOLAR_ACADEMIC=1 to enable)

Recommended Workflow

For best analysis quality, follow this order:

1. Download the paper: download_paper("2510.24701", title="Paper Title", domain="LLM")

2. Extract images: extract_paper_images("2510.24701") (auto-detects local PDF)

3. Deep analysis: analyze_paper(paper_json) (auto-detects local PDF, extracts full text)

Tip: Downloading the PDF before analysis enables full-text extraction, producing high-quality notes with specific data, formulas, and experimental results. Without a local PDF, analysis relies on the abstract only.

Configuration

.scholar.json

The .scholar.json file configures knowledge paths and academic research settings. See .scholar.example.json for a full example with comments.

Key sections:

• knowledge_dir — Path to knowledge cards directory

• index_path — Path to BM25 search index

• academic.research_interests — Your research domains, keywords, and arXiv categories

• academic.scoring — Paper scoring weights and dimensions

Environment Variables

Copy .env.example to .env and configure:

Project Structure

scholar-agent/
├── mcp_server.py              # MCP server (13 tools)
├── setup_mcp.py               # Embed into existing projects
├── pyproject.toml             # Package configuration
├── .scholar.json               # Project & academic configuration
├── schemas/                   # Answer + evidence JSON schemas
├── scripts/
│   ├── academic/              # Academic research modules
│   │   ├── arxiv_search.py    # arXiv + Semantic Scholar search
│   │   ├── conf_search.py     # Conference paper search (DBLP)
│   │   ├── paper_analyzer.py  # Deep-analysis note generation
│   │   ├── scoring.py         # 4-dim paper scoring engine
│   │   ├── image_extractor.py # Figure extraction from PDFs
│   │   ├── note_linker.py     # Wiki-link discovery + keyword linking
│   │   └── daily_workflow.py  # Daily recommendation pipeline
│   ├── scholar_config.py       # Configuration reader
│   ├── local_index.py         # BM25 index builder
│   ├── local_retrieve.py      # Knowledge retrieval
│   ├── close_knowledge_loop.py # Knowledge card builder
│   └── ...                    # Research, synthesis, governance, graph
├── knowledge/                 # Knowledge cards (gitignored, user-generated)
├── indexes/                   # Generated indexes (gitignored)
└── tests/                     # 247 tests

More Features

• Multi-perspective research — Parallel research from 5 perspectives (academic, technical, applied, contrarian, historical)

• Obsidian compatible — Standard Markdown + YAML frontmatter + [[wiki-links]]

• Knowledge governance CLI — Validate frontmatter, detect orphaned cards, find duplicates, manage lifecycle

• Provider fault tolerance — Each search source fails independently; falls back to local retrieval when offline

Testing

python -m pytest tests/ -v

247 tests, ~13s. No external services needed.

License

MIT — see LICENSE.