Awesome Genomic Skills

A curated list of skills and MCP servers for working with AI coding agents (Claude Code, GitHub Copilot, Codex, Cursor, Gemini CLI, etc.) in genomics and bioinformatics, alongside other useful repositories such as benchmarks and general AI coding skills collections.

What is a Skill and what is an MCP?

Skill: a Markdown file (plus optional scripts) that teaches an agent how to do a task — procedural know-how loaded into context on demand.

MCP server: a running service that gives an agent a connection to external systems (databases, tools, pipelines) via standardized tool calls.

Contents

• Awesome Genomic Skills
  • What is a Skill and what is an MCP?
  • Contents
  • Bioinformatics and Genomics Agent Skills
  • MCP Servers for Life Sciences
  • Benchmarks
  • General AI Coding Agent Skill Collections
  • Other Notable Awesome Lists!
  • Contributing

Bioinformatics and Genomics Agent Skills

Skill libraries and tool collections specifically targeting genomics, bioinformatics, and life sciences work with AI coding agents.

• science-skills
  • Description: Collection of ~36 agent skills spanning genomics, structural biology, cheminformatics, and literature search; wraps AlphaGenome (single-variant effect prediction), AlphaFold DB, and 30+ databases/tools (UniProt, Ensembl, gnomAD, GTEx, ClinVar, dbSNP, ChEMBL, PubChem, PDB, Foldseek, JASPAR, Reactome, STRING, Open Targets, Human Protein Atlas, PyMOL) for grounded, token-efficient scientific workflows. Apache-2.0; built for Google Antigravity but installable into any agent via npx skills add. Technical report.
  • Developers: Google DeepMind.
• openai/plugins — life-science-research
  • Description: OpenAI's Life Sciences research plugin for Codex; bundles 50 modular skills spanning human genetics, functional genomics, expression, pathway biology, protein structure, chemistry, and clinical evidence, wrapping 50+ public databases/tools (Ensembl, UniProt, gnomAD, GTEx, ClinVar, GWAS Catalog, Open Targets, ChEMBL, PubChem, RCSB PDB, AlphaFold, Reactome, STRING, Human Protein Atlas, cBioPortal, CellxGene, ENCODE, NCBI Entrez/BLAST/Datasets, and more). Works with mainline GPT-5.4 models, with GPT-Rosalind (trusted-access) for deeper reasoning. Closely parallels DeepMind's science-skills.
  • Developers: OpenAI.
• anthropics/life-sciences
  • Description: Anthropic's "Claude for Life Sciences" Claude Code Marketplace — a hybrid bundle rather than a pure skills library. It mixes first-party procedural/method skills (single-cell-rna-qc, scvi-tools, nextflow-development, clinical-trial-protocol-skill, scientific-problem-selection) with a large set of partner MCP servers and data integrations (10x Genomics, ChEMBL, Open Targets, PubMed, bioRxiv, Synapse, ToolUniverse, BioRender, Medidata, Cortellis, Owkin, Consensus, and more). Note that many entries are MCP servers, not skills.
  • Developers: Anthropic (with commercial life-sciences partners).
• sample-kiro-power-life-sciences
  • Description: AWS sample bundle for the Kiro IDE: 24 MCP servers wrapping 100+ databases/tools across genomics, proteomics, structural biology, and clinical/pharma (NCBI, Ensembl, ClinVar, gnomAD, UniProt, STRING, PDB, AlphaFold, ChEMBL, Open Targets, etc.), plus 10 domain skills and 16 workflows, with cross-database search and AWS HealthOmics pipeline execution. MIT-0; the MCP servers use standard MCP and are portable, though the skills and workflows are built for Kiro. Blog post.
  • Developers: AWS (AWS Samples).
• ClawBio
  • Description: Bioinformatics-native AI agent skill library; 95 reproducible, local-first skills for genomics tasks (variant calling, RNA-seq, population genetics) that work with Claude Code, Copilot, Codex, and other agents. Since v0.6.1 the whole library is also callable as an MCP server (uvx --from 'clawbio[mcp]' clawbio mcp) from Cursor, Claude Desktop, VS Code, or Zed (see the MCP section below). Skills are actively benchmarked.
  • Developers: Independent open-source project built on OpenClaw.
• SciAgent-skills
  • Description: 197 open-source skills for Claude Code, Cursor, Codex, and Windsurf, covering genomics-bioinformatics, proteomics-protein engineering, structural biology, drug discovery, systems biology, biostatistics, and scientific writing; achieves 92% accuracy on BixBench-Verified-50 (+26.7 pts over Claude Code baseline). The hosted OmicsHorizon web platform runs these skills in-browser.
  • Developers: The team behind the OmicsHorizon platform (Jaechang Lim).
• scientific-agent-skills
  • Description: Currently 135 skills covering various scientific areas including genomics, but also broader scientific areas like geospatial science etc.
  • Developers: K-Dense AI — an MIT-founded startup (Accel, Accel Atoms, Google AI Futures Fund) building an AI research platform (k-dense.ai); this is one of its open-source spin-offs.
• bioSkills
  • Description: SKILL.md files for bioinformatics with Claude Code, covering end-to-end pipelines like RNA-seq, variants, ChIP-seq, scRNA-seq, spatial, Hi-C, proteomics, microbiome, CRISPR, metabolomics, multi-omics, immunotherapy, outbreak analysis, and Mendelian randomization tools.
  • Developers: GPTomics (gptomics.com), an independent research lab working at the bioinformatics/AI intersection.
• Clair-skills
  • Description: Agent skill for the Clair suite of variant callers (Clair3, ClairS, Clair3-RNA, Clair-Mosaic); provides intelligent model selection, command generation, and troubleshooting for germline, somatic, mosaic, and RNA-seq variant calling from long-read and short-read sequencing data.
  • Developers: Ruibang Luo's BioAI Lab at the University of Hong Kong (HKU-BAL).
• ToolUniverse
  • Description: Ecosystem for building AI scientist systems; integrates 1,000+ machine learning models, datasets, and scientific APIs for data analysis, knowledge retrieval, and experimental design in biomedicine, with 68 pre-built agent skills covering drug discovery, precision oncology, and rare-disease diagnosis.
  • Developers: Harvard's Zitnik Lab (AI for Medicine and Science).
• operon
  • Description: AI-powered bioinformatics IDE bundling 180+ SKILL.md-format analysis protocols covering RNA-seq, scRNA-seq, ATAC-seq, ChIP-seq, WGS/WES, spatial transcriptomics, proteomics, GWAS, and external database query patterns (PubMed, GEO, GTEx, KEGG, UniProt, JASPAR, AlphaFold). The desktop app wraps Claude Code and adds HPC/SSH integration; the protocol files themselves are MIT-licensed Markdown in the repo's protocols/ directory, extractable for use with any SKILL.md-compatible agent.
  • Developers: Swarup Lab (UC Irvine).
• OpenClaw-Medical-Skills
  • Description: Meta-aggregation of 872 skills curated from 12+ upstream repositories, including ClawBio, ToolUniverse, GPTomics bioSkills, BioOS, and others; spans clinical workflows, genomics, drug discovery, bioinformatics pipelines, and medical device regulatory frameworks. Expect overlap with the upstream repos listed separately.
  • Developers: FreedomIntelligence, the medical-NLP research group at CUHK-Shenzhen / Shenzhen Research Institute of Big Data (led by Benyou Wang; also behind HuatuoGPT).

MCP Servers for Life Sciences

Model Context Protocol (MCP) servers that give AI agents direct access to bioinformatics databases, tools, and analysis pipelines.

• ChatSpatial - MCP server for spatial transcriptomics analysis through natural language; supports Scanpy, Squidpy, cell communication analysis, and spatial domain identification.
• biomcp - Single MCP server able to query multiple information sources, including clinical trials, genetic data & published medical literature.
• gget-mcp - MCP server wrapping the Pachter Lab gget bioinformatics toolkit. Exposes 13 tools covering gene search and metadata (Ensembl), sequence retrieval, BLAST/BLAT/MUSCLE alignment, expression data (ARCHS4), functional enrichment (Enrichr), protein structure (PDB, AlphaFold), cancer mutations (COSMIC), and single-cell queries (CellxGene).
• Seqera MCP - Hosted MCP server from Seqera Labs (the developers of Nextflow) exposing the Seqera Platform (workflow launch/management), Wave (container provisioning), nf-core modules, and SRA/ENA/GEO retrieval.
• knowledgebase-mcp - BioContextAI Knowledgebase MCP server, included in the BioContextAI registry (below), one of the most comprehensive single MCP packages (wraps STRINGDb, Open Targets, Reactome, UniProt, HPA, KEGG, AlphaFold, Ensembl, ClinicalTrials.gov, bioRxiv, etc.)
• ClawBio MCP - MCP mode of the ClawBio skill library (0.6.1): exposes all 95 genomics skills as callable tools over local stdio, letting different agents (Cursor, Zed, etc.) run analyses (variant calling, RNA-seq, population genetics).
• roda-mcp - AWS Labs MCP server for discovering and exploring datasets in the Registry of Open Data on AWS (RODA), covering 1,100+ public datasets across life sciences, climate, geospatial, and satellite imagery. Search/filter by keyword, organization, or license; inspect dataset details; and browse or sample public S3 bucket contents directly (no AWS account required) without downloading full files. Not genomics-specific, but useful for locating and previewing life-sciences datasets (e.g. SG-NEx) hosted on AWS.
• plant-genomics-mcp - Plant-genomics MCP server exposing 50 tools across 23 backends, keyed on TAIR-style loci with organism resolution across 12 crop and model species. Covers plant-specific resources that general bioinformatics servers do not (Ensembl Plants, Phytozome, Gramene, Planteome PO/TO, PlantCyc/PMN, AraGWAS, 1001 Genomes, ThaleMine, BAR, JASPAR) alongside the usual UniProt/KEGG/STRING/AlphaFold/PDBe/InterPro/Europe PMC, plus cross-source synthesis tools that compose several backends into a single gene report.

Existing registries and lists of MCP servers:

• BioContextAI Registry - Community-curated catalogue of biomedical MCP servers, with submission criteria requiring biomedical focus, free academic access, OSI-approved open-source licenses, and MCP specification compliance. Ships a cookiecutter template for new servers and follows Schema.org ontologies for metadata. A community hub for agentic biomedical systems
• MCPmed - Reference MCP implementations (GEO, STRING, UCSC Cell Browser, PLSDB) plus a cookiecutter template and HTML "breadcrumbs" discovery mechanism for transitioning legacy services to MCP. Published as a call paper in Briefings in Bioinformatics. MCPmed: a call for Model Context Protocol-enabled bioinformatics web services for LLM-driven discovery
• awesome-mcp-servers - General-purpose MCP server list with a Biology, Medicine, and Bioinformatics subsection.

MCP related tools:

• BioinfoMCP - Not strictly an MCP, a converter that auto-generates MCP servers from existing tool documentation, plus a benchmark of the converted tools. Preprint available here

Benchmarks

Benchmarks, evaluations, and other helpful resources at the intersection of AI and genomics/bioinformatics.

Agent Capability Benchmarks:

• BioAgent Bench - 10 end-to-end bioinformatics pipeline tasks (RNA-seq, variant calling, metagenomics, single-cell, transcript quantification, etc.) with concrete output artifacts; includes a perturbation suite (corrupted inputs, decoy reference files, prompt bloat) — probes agent robustness under controlled stress and shows that correct high-level pipeline construction does not guarantee reliable step-level reasoning.
• BioMed-AQA - 327 open-ended biomedical analysis tasks across omics, visualisation, machine learning, statistics, and precision medicine; uses milestone-based grading against reference analytical steps, with a complementary 172-question multiple-choice subset. Released alongside the BioMedAgent system in Nature Biomedical Engineering — same-team caveat applies to headline scores.
• BiomniBench - Process-level evaluation framework: 100 biomedical data-analysis tasks curated from high-impact papers by original authors or domain experts; grades the full agent trajectory (reasoning trace + final answer) against expert-designed rubrics via an LLM judge — addresses the outcome-only blind spots in benchmarks like BixBench.
• BixBench - Comprehensive benchmark for LLM-based agents on real-world computational biology tasks; tests agents' ability to explore biological datasets, perform multi-step analyses, and interpret results — useful for evaluating which agents and skills perform best on genomics work.
• CompBioBench - Genentech-released benchmark of 100 computational biology questions spanning single-cell, epigenomics, genomics, transcriptomics, human genetics, and ML; agents start from a bare-minimum environment and must fetch their own tools and data, with exact-string-match grading on a single ground-truth answer.
• LAB-Bench - 2,400+ multiple-choice questions across 8 subtasks of practical biology research (literature search, figure/table interpretation, database access, wet-lab protocols, sequence analysis, cloning scenarios); FutureHouse's predecessor to BixBench, probing biological knowledge and reasoning rather than agentic execution.

Skills benchmarks:

• SkillsBench - General-purpose benchmark for measuring whether agent skills actually help: 86 tasks across 11 domains (including healthcare), each run under three conditions — no skills, curated skills, and self-generated skills — with deterministic verifiers. Across 7,308 trajectories, curated skills lifted average pass rate by +16.2 pts (ranging from +4.5 pts for software engineering to +51.9 pts for healthcare), while self-generated skills gave no average benefit and focused 2–3 module skills beat comprehensive documentation. Not bioinformatics-specific, but the canonical "do skills actually work" benchmark. Paper.

General AI Coding Agent Skill Collections

Popular repositories of reusable skills for AI coding agents. Each entry is useful for genomics and bioinformatics coding work; descriptions explain how.

Skills specs:

• agentskills - Official specification and documentation for the Agent Skills standard; defines the cross-platform format used by skills in this list and on Claude Code, Copilot, Codex, Cursor, and Gemini CLI.

General skills collections (non-exhaustive):

• superpowers - Complete software development methodology and skills framework for coding agents; enforces TDD, spec-driven design, and subagent-driven development — practices that improve reproducibility and correctness in complex genomics pipelines.
• anthropics/skills - Official Anthropic reference implementation and specification for Claude agent skills; the canonical starting point for building custom skills that teach Claude how to handle bioinformatics tasks and lab workflows.
• openai/skills - OpenAI's official Skills Catalog for Codex; ships system, curated, and experimental agent skills installable via the in-Codex $skill-installer, built on the cross-platform Agent Skills open standard. The OpenAI counterpart to anthropics/skills, and a useful reference for packaging reusable, repeatable bioinformatics coding workflows for Codex.
• andrej-karpathy-skills - Single-file coding guidelines derived from Andrej Karpathy's observations on LLM pitfalls; instils simplicity, surgical changes, and goal-driven execution — critical disciplines when AI agents write or modify genomics analysis code.
• awesome-copilot - Community-curated instructions, agents, skills, and configurations for GitHub Copilot, including prompt templates applicable to scientific and data-analysis workflows.
• awesome-llm-skills - Curated list of LLM and AI agent skills, resources, and tools for customizing AI agent workflows across Claude Code, Codex, Gemini CLI, and custom agents.

Other Notable Awesome Lists!

• Awesome-Scientific-Skills - An open, curated collection of agent skills for scientific research spanning bioinformatics, cheminformatics, data analysis, scientific writing, and literature search; currently a curated link collection, with plans to consolidate selected skills into a unified, clone-ready repository. Developed by InternScience, the open-source hub of the AI for Science Center at Shanghai AI Laboratory, which open-sources agents, LLMs/MLLMs, tools, and datasets to accelerate scientific discovery across disciplines.

Contributing

Contributions are welcome. Please read the contribution guidelines before submitting a pull request.