Content
# Claude Science — Reference Extraction
This repository contains the extracted agents, skills, MCP server sources, system prompt sections, and architecture documentation from the **Claude Science** macOS application (bundle ID `com.anthropic.operon`, version `0.1.0-dev.20260630`).
Claude Science is Anthropic's scientific computing research assistant — a Bun-compiled macOS app that serves a local web UI (at `localhost:8765`) backed by a multi-agent Claude harness with 30 bundled skills and 24 science-domain MCP servers.
## Contents
```
agents/ # Agent configs (metadata.yaml) for all 4 bundled agents
operon/ # Root general-purpose agent ("Claude Science")
reviewer/ # Transcript hallucination/fabrication reviewer
bookmarker/ # Session breadcrumb extractor
onboarding/ # First-run onboarding interview host
skills/ # 29 bundled skills (SKILL.md + supporting assets)
alphafold2/ # AlphaFold2 protein structure prediction
boltz/ # Boltz-1 structure prediction
borzoi/ # Borzoi genomic sequence model
chai1/ # Chai-1 biomolecular structure prediction
compute-env-setup/ # Setting up compute on SSH/Slurm/Modal/cloud
customize/ # Agent profile + skill CRUD via host SDK
diffdock/ # DiffDock molecular docking
esmfold2/ # ESMFold2 protein structure
evo2/ # Evo2 genomic foundation model
fair-esm2/ # FAIR ESM2 protein language model
figure-composer/ # Multi-panel scientific figure layout
figure-style/ # Publication-grade figure correctness rules
indication-dossier/ # Therapeutic indication research dossier
ligandmpnn/ # LigandMPNN protein-ligand design
literature-review/ # Structured scientific literature review
managed-model-endpoints/ # Using Anthropic-hosted inference endpoints
openfold3/ # OpenFold3 structure prediction
paper-narrative/ # Whole-paper figure arc planning
pdf-explore/ # PDF document exploration
product-self-knowledge/ # App capabilities and feature reference
proteinmpnn/ # ProteinMPNN protein sequence design
remote-compute-modal/ # GPU jobs via Modal (byoc:modal)
remote-compute-ssh/ # Jobs on SSH/SLURM clusters
scgpt/ # scGPT single-cell foundation model
scvi-tools/ # scVI-tools single-cell analysis
self-awareness/ # Claude Science's own SQLite DB schema + SDK surface
skill-creator/ # Author new skills iteratively
solublempnn/ # SolubleMPNN solubility-optimized design
using-model-endpoint/ # Using custom model API endpoints
mcp-servers/ # Bundled MCP server source code
bio-tools/ # Python launcher + 24 science-domain servers
run_server.py # Launcher — dispatches to lib/mcp_* packages
lib/
mcp_pubmed/ # PubMed literature search and fetch
mcp_bio/ # General biology databases
mcp_biomart/ # Ensembl BioMart gene/transcript queries
mcp_biorxiv/ # bioRxiv/medRxiv preprint access
mcp_cancer_models/ # Cancer model databases
mcp_cellguide/ # CellxGene cell type guide
mcp_chembl/ # ChEMBL bioactivity and drug data
mcp_chemistry/ # Chemical structure and properties
mcp_clinical_genomics/ # Clinical genomics databases
mcp_clinical_trials/ # ClinicalTrials.gov access
mcp_drug_regulatory/ # FDA, EMA regulatory data
mcp_expression/ # Gene expression databases (GEO, etc.)
mcp_genes_ontologies/ # Gene Ontology and pathway databases
mcp_genomes/ # Genome databases (NCBI, Ensembl)
mcp_human_genetics/ # GWAS, gnomAD, ClinVar
mcp_literature/ # Broad literature access (Europe PMC etc.)
mcp_omics_archives/ # SRA, GEO, ENA bulk omics archives
mcp_protein_annotation/ # UniProt, InterPro annotation
mcp_regulation/ # Regulatory genomics (ENCODE, JASPAR)
mcp_research_resources/ # Research resource databases
mcp_rna/ # RNA databases (miRBase, RNAcentral)
mcp_servers_common/ # Shared server utilities
mcp_structures_interactions/ # PDB, protein-protein interactions
mcp_variants/ # Variant databases (ClinVar, gnomAD)
mcp_zinc/ # ZINC compound database
ketcher-chemistry/ # Ketcher chemical structure editor (Node.js widget)
system-prompts/ # Extracted system prompt sections (RULES_*)
RULES_CORE.md
RULES_SECURITY.md
RULES_BIOSECURITY.md
RULES_CODE_EXECUTION.md
RULES_NETWORK_SANDBOX.md
RULES_CAPABILITY_GUIDANCE.md
RULES_PERSONAL_HEALTH.md
RULES_SECURITY_SANDBOX.md
docs/
architecture.md # Full system architecture
architecture.html # Interactive visual diagram of the full architecture
agent-system.md # Agent hierarchy, trust model, reviewer/bookmarker
skill-system.md # Skill format, search, loading, authoring
mcp-integration.md # MCP server patterns, tool call flow
compute-environments.md # SSH, SLURM, Modal, HPC — full guide
artifact-system.md # Artifact save/embed/provenance
biosecurity.md # Biosecurity screener design
sqlite-schema.md # Internal metadata database schema
scheduler.md # Scheduled agent execution (routine_schedules)
managed-endpoints.md # Skill-managed local processes
marketplace.md # External skill sources + license assents
reuse-guide.md # How to adapt these patterns for your own agent
drizzle/sqlite/ # 95 Drizzle ORM migration files — authoritative DB schema source
examples/ # 4 complete real Claude Science sessions with full artifacts
crispr_screen/ # Genome-wide kinome CRISPR knockout library design
enzyme_engineering/ # IS621 bridge recombinase engineering
extremophile/ # NIF3 ancestral sequence reconstruction
immunotherapy/ # Sade-Feldman melanoma scRNA-seq analysis
```
## Quick Start — Reuse Guide
See [docs/reuse-guide.md](docs/reuse-guide.md) for a practical walkthrough of adapting this agent/skill/MCP pattern for your own Claude-based application.
## Architecture Diagram
[docs/architecture.html](docs/architecture.html) is a self-contained interactive diagram of the full system — process layout, agent hierarchy, system prompt assembly, kernel architecture, MCP server grid, SQLite schema, compute backends, and the biosecurity/verification pipeline. Open it directly in a browser, or view the hosted copy:
**https://docs-internal.formbio.com/formbio/claude-science-reference/architecture.html**
## License
The skill and agent configuration files in this repository are extracted from the Claude Science application. System prompt content and skill content are copyright Anthropic, PBC. Third-party licenses for the MCP server dependencies are in `skills/THIRD_PARTY_LICENSES.md`.
MCP Config
Below is the configuration for this MCP Server. You can copy it directly to Cursor or other MCP clients.
mcp.json
Connection Info
You Might Also Like
ScienceClaw
🔬🦞 A self-evolving AI research colleague for scientists. 285 skills, 25+...
ScienceClaw
ScienceClaw is a personal research assistant built with LangChain DeepAgents...
garmin-connect-mcp
MCP server for Garmin Connect — access 61 health, fitness and activity tools...
prism-coder
The Mind Palace for AI Agents - HIPAA-hardened Cognitive Architecture with...
garmin-givemydata
It's YOUR data. Take it back. Get your Garmin Connect health data into a...
whoop-mcp
MCP server to connect to whoop API