claude-scientific-writer

Claude Scientific Writer

🚀 Looking for more advanced capabilities? For end-to-end scientific writing, deep scientific search, advanced image generation and enterprise solutions, visit www.k-dense.ai

Stay up to date: Read the K-Dense blog, and follow us on X, LinkedIn, and YouTube for new features, release announcements, walkthroughs, research workflow demos, and examples you can use with Scientific Writer. See From the K-Dense Blog for posts that map to this tool.

🎬 Prefer to watch first? The Open Source, Locally-Hosted Scientific Writer Workflow shows the writer researching and drafting a paper end to end. More walkthroughs on the K-Dense YouTube channel.

A deep research and writing tool that combines the power of AI-driven deep research with well-formatted written outputs. Generate publication-ready scientific papers, reports, posters, grant proposals, literature reviews, and more academic documents—all backed by real-time literature search and verified citations.

Scientific Writer performs comprehensive research before writing, ensuring every claim is supported by real, verifiable sources. Features include real-time research lookup through Parallel Search and Extract, intelligent paper detection, comprehensive document conversion, and AI-powered diagram generation. You can use it as a Claude Code plugin, Python package, or native CLI. The announcement post, Claude Scientific Writer: Our Open Source Tool for AI-Powered Research Writing, covers why we built it and what it produces.

Quick Start

Prerequisites

• Python 3.10 or newer
• ANTHROPIC_API_KEY (required)
• Parallel CLI authentication (parallel-cli login) or PARALLEL_API_KEY
• OPENROUTER_API_KEY (optional, for AI image generation: schematics, figures, slides, and infographics)
• A LaTeX distribution for PDF generation (pdflatex, bibtex, and preferably latexmk)
• Optional: LibreOffice for Office rendering/recalculation and FFmpeg for media conversion

Installation Options

Option 1: Claude Code Plugin (Recommended) ⭐

The easiest way to use Scientific Writer is as a Claude Code plugin. See the Plugin Installation section below.

Option 2: Install from PyPI (CLI/API Usage)

pip install scientific-writer

# Optional bundled-script runtimes
pip install "scientific-writer[analysis]"  # cohort statistics and survival analysis
pip install "scientific-writer[office]"    # DOCX/PPTX/XLSX and MarkItDown helpers

Option 3: Install from source with uv

git clone https://github.com/K-Dense-AI/claude-scientific-writer.git
cd claude-scientific-writer
uv sync

Configure API keys

# .env file (recommended) — see .env.example for the full list
echo "ANTHROPIC_API_KEY=your_key" > .env
echo "PARALLEL_API_KEY=your_parallel_key" >> .env            # research lookup and web search
echo "OPENROUTER_API_KEY=your_openrouter_key" >> .env        # optional: AI image generation
# or export in your shell
export ANTHROPIC_API_KEY='your_key'

Install and authenticate the research CLI when it is not already available:

uv tool install "parallel-web-tools[cli]==0.7.1"
parallel-cli login
parallel-cli auth

Usage Options

Use as Plugin (Recommended)

After installing the plugin and running /claude-scientific-writer:scientific-writer-init, simply ask Claude:

> Create a Nature paper on CRISPR gene editing. Present experimental_data.csv 
  (efficiency across 5 cell lines), include Western_blot.png and flow_cytometry.png 
  showing 87% editing efficiency (p<0.001). Compare with literature benchmarks.

> Generate an NSF grant proposal presenting preliminary data from quantum_results.csv 
  (99.2% gate fidelity), circuit_topology.png, and error_rates.csv. 
  Include 5-year timeline with milestones_budget.xlsx.

> @research-lookup Find papers on mRNA vaccine efficacy (2022-2024). Compare 
  with our trial_outcomes.csv (n=500, 94% efficacy) and antibody_titers.png.

Use the CLI

# If installed via pip
scientific-writer
scientific-writer --effort high

# If installed from source with uv
uv run scientific-writer

Use scientific-writer --help for permission, budget, token-usage, and input-consumption controls. Input files are preserved by default; --consume-inputs explicitly removes them after a successful copy.

Use the Python API

import asyncio
from scientific_writer import generate_paper

async def main():
    # Detailed prompt with specific data and figures
    async for update in generate_paper(
        query=(
            "Create a Nature paper on CRISPR gene editing. "
            "Present editing_efficiency.csv (5 cell lines, n=200 cells each). "
            "Include Western blot (protein_knockout.png) showing target depletion, "
            "flow cytometry data (editing_percentages.png) with 87% efficiency in HEK293, "
            "and off_target_analysis.csv showing <0.1% off-target effects. "
            "Compare results to published Cas9 benchmarks (typically 70-75% efficiency)."
        ),
        data_files=[
            "editing_efficiency.csv",
            "protein_knockout.png",
            "editing_percentages.png",
            "off_target_analysis.csv"
        ]
    ):
        if update["type"] == "text":
            print(update["content"], end="", flush=True)
        elif update["type"] == "progress":
            print(f"[{update['stage']}] {update['message']}")
        elif update["type"] == "result":
            print(f"✓ PDF: {update['files']['pdf_final']}")
            print(f"  Figures: {len(update['files']['figures'])} included")

asyncio.run(main())

🎯 Use as a Claude Code Plugin (Recommended)

Scientific Writer works best as a Claude Code (Cursor) plugin, providing seamless access to all scientific writing capabilities directly in your IDE. No CLI required!

Quick Start - Plugin Installation

1. Add the plugin marketplace in Claude Code:

   /plugin marketplace add https://github.com/K-Dense-AI/claude-scientific-writer
   

2. Install the plugin:

   /plugin install claude-scientific-writer
   

3. Restart Claude Code when prompted.

4. Initialize in your project:

   /claude-scientific-writer:scientific-writer-init
   

   This creates a CLAUDE.md file with comprehensive scientific writing instructions and makes all 26 selected skills available.

5. Start using immediately:

   # Create papers with data and figures
   > Create a Nature paper on CRISPR gene editing. Present knockout_efficiency.csv 
     (5 cell lines tested), include Western blot (protein_levels.png) and flow 
     cytometry data (editing_rates.png). Highlight 87% efficiency in HEK293 cells.
   
   > Write an NSF grant proposal for quantum computing. Present preliminary results 
     from gate_fidelity.csv (99.2% fidelity), include circuit_diagram.png and 
     error_analysis.png. Compare to state-of-art 95% baseline.
   
   > Generate conference poster. Feature results from clinical_trial.csv 
     (n=150), survival_curves.png, biomarker_heatmap.png, and mechanism_diagram.svg.
   
   # Use specific skills with research data
   > @research-lookup Find papers on mRNA vaccine efficacy (2022-2024). Compare 
     with our trial_data.csv showing 94% efficacy and antibody_titers.xlsx.
   
   > @peer-review Evaluate this manuscript. Reference sample size in methods.csv 
     (n=30) and effect_sizes.png. Assess if statistical power is adequate.
   
   > @clinical-reports Create case report for autoimmune disorder. Include patient_labs.xlsx 
     (6 months data), MRI_scans/ folder, treatment_timeline.csv showing response.
   

Why Use the Plugin?

• ✅ No CLI Required - Everything works directly in Claude Code
• ✅ Instant Access - All 26 selected skills available immediately
• ✅ IDE Integration - Files created and edited in your project
• ✅ Context Aware - Skills understand your project structure
• ✅ Seamless Workflow - No switching between tools

Available Skills

When installed as a plugin, you get instant access to a reproducible subset of K-Dense-AI/scientific-agent-skills, pinned in skills.lock.json. Background on the skills model is in Agent Skills: The Final Piece for AI-Powered Scientific Research and K-Dense Web vs Scientific Agent Skills. Read Security in the Science Agent Era before installing skills in a lab environment.

• scientific-schematics - AI diagram generation with Nano Banana 2 (CONSORT, neural networks, pathways); see Benchmarking Nano Banana 2 Lite for Scientific Image Generation
• research-lookup - Real-time literature search
• peer-review - Systematic manuscript evaluation
• citation-management - BibTeX and reference handling
• clinical-reports - Medical documentation standards
• research-grants - NSF, NIH, DOE proposal support
• scientific-slides - Research presentations
• latex-posters - Conference poster generation
• hypothesis-generation - Scientific hypothesis development
• market-research-reports - Comprehensive 50+ page market analysis reports with visuals
• And 16 more specialized skills...

For local plugin development and testing, see the Development Guide.

Features

📝 Document Generation

• Scientific papers with IMRaD structure (Nature, Science, NeurIPS, etc.)
• Clinical reports (case reports, diagnostic reports, trial reports, patient documentation)
• Research posters using LaTeX (beamerposter, tikzposter, baposter)
• Grant proposals (NSF, NIH, DOE, DARPA) with agency-specific formatting
• Literature reviews with systematic citation management
• Scientific schematics powered by Nano Banana 2 (CONSORT diagrams, neural architectures, biological pathways, circuit diagrams) — image-model benchmark

🤖 AI-Powered Capabilities

• Real-time research lookup powered by Parallel Search and Extract; why verification matters is covered in The AI Co-Scientist Is Here. The Bottleneck Is Verification.