paper2agent

name	paper2agent
description	Convert scientific papers into executable computational workflows and MCP tools
metadata	{"openclaw":{"requires":{"env":["OPENAI_API_KEY"],"bins":["python3"]},"primaryEnv":"OPENAI_API_KEY"}}

Paper2Agent — Literature-to-Tool Conversion

Extract computational methods from scientific papers and generate executable MCP tool specifications with verifiable reward criteria.

Capabilities

• Method extraction: Parse papers to identify computational methodologies, input parameters, algorithm steps, and validation criteria
• Reward criteria extraction: Derive verifiable reward criteria (thermostability, structural accuracy, binding affinity, conservation) from paper content
• MCP tool generation: Generate MCP tool specifications with typed input/output schemas
• Script generation: Optionally generate runnable Python tool skeletons
• Artifact DAG integration: Record paper, methodology, and tool artifacts with full provenance tracking (Layer 3)
• Domain classification: Automatically classify papers into MD, structural, bioinformatics, or general domains

Usage

Provide a paper (PDF, text file, or raw text). The skill extracts computational methods and generates executable workflow configurations or scripts.

# Generate tool configs from a paper
python skills/paper2agent/scripts/run_paper2agent.py paper.txt -o output/

# Generate runnable scripts with artifact tracking
python skills/paper2agent/scripts/run_paper2agent.py paper.txt \
    --output-format script \
    --artifact-root .artifact_store

# Extract a specific method only
python skills/paper2agent/scripts/run_paper2agent.py paper.txt \
    --target-method stability_prediction

Parameters

• paper_source: Path to paper (PDF/text file) or raw text string
• output_dir: Directory for output files (default: output/paper2agent)
• artifact_root: Root directory for artifact DAG storage (optional — enables provenance)
• target_method: Extract only a specific methodology (optional — extracts all if omitted)
• output_format: Output format — config (JSON specs, default) or script (runnable Python)

Pipeline

1. Load paper — read PDF/text file or accept raw text
2. Classify domain — MD, structural biology, bioinformatics, or general
3. Extract methodologies — identify computational methods with inputs, outputs, algorithm steps
4. Extract reward criteria — derive verifiable evaluation criteria from paper content
5. Generate MCP tool specs — create typed tool specifications with confidence scores
6. Record artifacts — store paper, methodology, and tool artifacts in the DAG with parent lineage
7. Write outputs — JSON summary + optional generated Python scripts

Supported Domains

Domain	Example Methods
Molecular Dynamics	MD simulation protocols, trajectory analysis
Structural Biology	Structure prediction, binding analysis
Bioinformatics	Conservation analysis, sequence alignment
Stability	Mutation stability prediction, thermostability