Bioinformatics Agent Skills

By Pipette.bio — Agentic bioinformatics for wet-lab biologists.

Important: The Skill Graph front end server and this repo are maintained by Claude Opus 4.7 (somewhat autonomously).

A hosted MCP server that gives Claude Code (and any MCP client) direct access to a knowledge graph of 78 bioinformatics workflows — single-cell, variant calling, RNA-seq, drug discovery, metagenomics, and more.

Query pipelines mid-conversation. No install. No data download. Just point your AI coding agent at a URL.

🔗 Live graph explorer: skillgraph.pipette.bio 🔗 Main site: pipette.bio

Quick start

1. Add the MCP server to Claude Code

claude mcp add --transport http --scope user skillgraph https://skillgraph.pipette.bio/mcp

Or edit ~/.claude.json manually:

{
  "mcpServers": {
    "skillgraph": {
      "type": "http",
      "url": "https://skillgraph.pipette.bio/mcp"
    }
  }
}

2. Restart Claude Code

MCP servers only load at startup. Fully exit and reopen your session.

3. Verify it's connected

claude mcp list

You should see:

skillgraph: https://skillgraph.pipette.bio/mcp (HTTP) - ✓ Connected

4. Ask Claude anything

> What tools does the scanpy skill use?
> Find the shortest path from wgs-alignment to pathway-enrichment
> List all single-cell skills
> What comes after variant-calling in a typical pipeline?
> Search for CRISPR-related skills

Claude will call the MCP tools automatically and use the results in its answer.

What's in the graph

• 78 bioinformatics skills — scanpy, seurat, variant-calling, chipseq, atacseq, gwas-analysis, admet-prediction, molecular-docking, metagenome-binning, and more
• 483 pipeline transitions — each annotated with the data types that flow between steps (BAM, VCF, H5AD, FASTQ, etc.), the number of supporting papers, and whether the edge was confirmed in a curated ground-truth graph
• ~300 tools across skills — Scanpy, Seurat, bwa, samtools, DESeq2, GATK, fastp, MultiQC, STAR, Salmon, Cell Ranger, AlphaFold, AutoDock, and many more
• Full SKILL.md documentation for each skill — triggers, required tools, input/output formats, workflow steps, critical rules, and output artifacts

Tools exposed by the MCP server

Tool	What it does
get_skill	Returns full SKILL.md documentation for a skill, plus overview, tools, I/O, and connections
list_skills	Lists all available skills, optionally filtered by domain keyword
search_skills	Keyword search across skill IDs, trigger keywords, and tool names
get_transitions	Upstream and/or downstream skills with edge evidence (paper counts, data types)
find_path	Shortest pipeline path between two skills, annotated with data flow
get_graph_stats	Summary statistics: total skills, edges, domain breakdown

Example prompts

Explore a specific skill:

What does the atacseq skill do? What tools does it need?

Build a multi-step workflow:

I have raw FASTQ files from a WGS experiment. What's the sequence of skills
I should run to get a list of pathway enrichments?

Understand the graph:

How many single-cell skills are in the graph? Give me a breakdown by domain.

Find related work:

Show me skills that produce VCF files as output. Which ones are most commonly
used downstream of variant-calling?

How it works

Claude Code
  │ HTTPS POST
  ▼
skillgraph.pipette.bio/mcp
  │ (CloudFront)
  ▼
API Gateway → AWS Lambda → in-memory graph + skill docs

Everything runs on our infrastructure. Your Claude Code client just speaks MCP over HTTPS. No install, no data download, no local dependencies.

About the data

The SkillGraph knowledge graph was built from open-access bioinformatics papers indexed in PubMed Central. Tool mentions were extracted with a fine-tuned PubMedBERT NER model, pipeline transitions were inferred from document-level tool ordering, and data-type compatibility was validated against the EDAM ontology. The full interactive graph explorer is at skillgraph.pipette.bio.

About us

Built by Variome Analytics / Pipette.bio. We build agentic bioinformatics for wet-lab biologists.

Issues and feedback

Found a bug? Want a new tool? Have a skill we should add to the graph? Open an issue.

License

MIT — see LICENSE.