ScienceStudio

An IDE for Research — Word + LaTeX + AI, local-first.

What is ScienceStudio?

A research-focused IDE that gives AI coding agents research superpowers. Instead of building another AI, we use Claude Code or OpenCode as the brain and build research tools (MCP servers) that any agent can use.

Supports both Word AND LaTeX — covering ~95% of academic researchers.

Architecture

┌─────────────────────────────────────────────────────┐
│  VS Code Extension                                  │
│  ├── OnlyOffice (.docx) ─┬─ Inline AI (Cmd+K)       │
│  ├── LaTeX Editor (.tex) ─┘                         │
│  ├── PDF Library                                    │
│  └── Focus Mode (hides VS Code complexity)          │
├─────────────────────────────────────────────────────┤
│  Choose Your Brain:                                 │
│  ┌─────────────┐  ┌─────────────┐                   │
│  │ Claude Code │  │  OpenCode   │                   │
│  │ (Claude)    │  │ GPT/Gemini/ │                   │
│  │ [Default]   │  │ LLaMA/etc   │                   │
│  └──────┬──────┘  └──────┬──────┘                   │
│         └────────┬───────┘                          │
└──────────────────┼──────────────────────────────────┘
                   │ MCP Protocol
                   ▼
┌─────────────────────────────────────────────────────┐
│  MCP Servers (Research Tools)                       │
│  ├── pdf-mcp: Read and extract from PDFs            │
│  ├── library-mcp: Vector search over your papers    │
│  ├── citation-mcp: Find papers, get BibTeX          │
│  └── docx-mcp: Read/write Word documents            │
└─────────────────────────────────────────────────────┘

Key Features

Inline AI Assistant (Cmd+K)

Select text in your document, press Cmd+K, and get research-powered AI:

Command	What it does
Refine	Polish your writing
Expand	Elaborate on the point
Add citations	Find supporting papers from your library
Find sources	Search Semantic Scholar for new papers
Verify claim	Check if your sources support the claim
Strengthen	Add evidence and improve reasoning

Two Editors, One Experience

Format	Editor	Users
.docx (Word)	OnlyOffice	Bio, Med, Social Sciences (~60-70%)
.tex (LaTeX)	Monaco	Math, Physics, CS (~10-30%, 70-90% in STEM)

Both editors share the same inline AI and MCP tools.

Choose Your AI Brain

Agent	Models	Best For
Claude Code (default)	Claude 3.5/4	Best agent orchestration
OpenCode	GPT-4, Gemini, LLaMA, 75+	User choice, open-source

Research MCP Servers

# Search your paper library
> library_search("cognitive load affects memory")
Found 3 papers: Smith 2023, Jones 2022, ...

# Look up a paper by DOI
> citation_lookup_doi("10.1038/nature12373")
Title: "Structural basis of...", Authors: ...

# Extract sections from a PDF
> pdf_extract_sections("/path/to/paper.pdf")
Abstract: "...", Methods: "...", Results: "..."

Why ScienceStudio?

Tool	Word	LaTeX	AI	Local-First
MS Word	✅	❌	✅ Copilot	❌
Google Docs	✅	❌	✅ Gemini	❌
Overleaf	❌	✅	❌	❌
ScienceStudio	✅	✅	✅	✅

Our differentiator: Only tool that supports BOTH Word AND LaTeX with AI + local-first + agent-agnostic.

Quick Start

1. Clone the repo

git clone https://github.com/AIScienceStudio/ScienceStudio.git
cd ScienceStudio

2. Install the VS Code extension

cd extensions/sciencestudio-core
npm install
npm run compile

3. Set up MCP servers

# Create Python environment
conda create -n sciencestudio python=3.12
conda activate sciencestudio

# Install MCP servers
cd mcp-servers/pdf-mcp && pip install -r requirements.txt
cd ../library-mcp && pip install -r requirements.txt
cd ../citation-mcp && pip install -r requirements.txt
cd ../docx-mcp && pip install -r requirements.txt

4. Register MCP servers with your agent

For Claude Code (~/.claude.json):

{
  "mcpServers": {
    "pdf": { "command": "python", "args": ["/path/to/mcp-servers/pdf-mcp/server.py"] },
    "library": { "command": "python", "args": ["/path/to/mcp-servers/library-mcp/server.py"] },
    "citation": { "command": "python", "args": ["/path/to/mcp-servers/citation-mcp/server.py"] },
    "docx": { "command": "python", "args": ["/path/to/mcp-servers/docx-mcp/server.py"] }
  }
}

For OpenCode — same format in ~/.config/opencode/opencode.json

5. Launch

code --extensionDevelopmentPath=/path/to/extensions/sciencestudio-core

Roadmap

Phase 1: Foundation (Current)

• Project architecture defined
• MCP servers implemented (pdf, library, citation, docx)
• VS Code extension skeleton
• OnlyOffice integration
• Inline AI assistant (Cmd+K)

Phase 2: Editors

• OnlyOffice WebView for .docx
• LaTeX editor with PDF preview
• PDF viewer with annotations

Phase 3: Intelligence

• Full inline AI with MCP tools
• Citation autocomplete
• Claim verification
• Smart search across library

Phase 4: Polish

• Focus Mode (hide VS Code complexity)
• One-click installer
• Cloud sync (optional)

Documentation

Doc	Description
CLAUDE.md	Project overview and setup
Design Choices	Architecture decisions
Market Research	Word vs LaTeX usage data
OnlyOffice Design	Editor integration plan
Tasks	Current development tasks

Project Structure

ScienceStudio/
├── extensions/
│   └── sciencestudio-core/     # VS Code extension
├── mcp-servers/
│   ├── pdf-mcp/                # PDF extraction
│   ├── library-mcp/            # Vector search (ChromaDB)
│   ├── citation-mcp/           # Semantic Scholar API
│   └── docx-mcp/               # Word document handling
├── docs/
│   ├── design-choices.md
│   ├── market-research.md
│   └── plans/
└── plan/
    └── tasks.md

Target Users

Primary: PhD students and academic researchers

• Psychology, Biology, Medicine (Word users)
• Math, Physics, CS (LaTeX users)

Use case: Manage 50-100+ PDFs, write papers, find evidence, verify citations.

Contributing

We're building the future of research tools. Contributions welcome!

• UI/UX — Make OnlyOffice/LaTeX integration seamless
• AI/ML — Improve MCP server capabilities
• Research — Help us understand researcher workflows

License

MIT — Open source forever.

---

Your data stays on your machine. Always.

Documentation · Issues · Discussions