Introducing biocontext7: Deep Skill Bundles for AI Bioinformatics Agents
Today we’re publicly launching biocontext7 — 2,000+ deep skill bundles for AI bioinformatics agents, served via MCP.
The Problem
Bioinformatics has an agent-reliability problem. There are thousands of tools spread across Bioconductor, PyPI, CRAN, Galaxy, and GitHub. When you ask a coding assistant to help with a workflow, it often hallucinates package names, suggests deprecated APIs, or conflates tools with similar names. The prompts that make it work are hand-crafted, fragile, and they rot as upstream tools change.
What biocontext7 Does
biocontext7 compiles structured, hash-verified documentation bundles from live upstream sources and serves them to agents via MCP:
- 2,000+ deep skill bundles across 13 domains (genomics, transcriptomics, single-cell, metagenomics, systems biology, clinical genomics, ML, imaging, workflows, population genetics, utilities, and more)
- SHA-256 provenance — every bundle is hash-verified and traceable to a specific upstream version
- Execution-certified docs — snippets are compiled from real sources, not generated
- Native MCP integration — Claude Code, Cursor, VS Code, JetBrains, and any MCP-compatible client
- No hand-crafted prompts, no hallucinated tool APIs — agents work from canonical documentation, not guesses
- Open access — no login, no registration, no payment, no tracking
Install
Add biocontext7 to Claude Code in one command:
claude mcp add biocontext7 -- npx @biocontext7/mcp
For Cursor, VS Code, or JetBrains, follow the client setup guide.
How It Works
biocontext7 exposes MCP tools that agents call at reasoning time:
resolve-library-id— look up a tool by name or keyword, returns matched IDs and metadataget-library-docs— fetch versioned documentation for a specific tool with topic filtering and token budget controlfind-skills— discover skill bundles for a workflow by capabilitysearch-and-get— one-shot lookup that returns compiled, executable guidance
Because the bundles are compiled from live upstream sources and verified with SHA-256, agents don’t hallucinate APIs — they work from documentation that was present in the tool’s actual release.
Why Deep Skill Bundles
A bundle is more than a README. Each skill is an executable contract: what it does, what inputs it expects, what outputs it produces, the exact command line or Python/R API, and citations back to the original tool. That structure is what lets an agent chain multiple tools reliably — a CRISPR guide design workflow, a variant annotation pipeline, a single-cell atlas embedding — without inventing steps.
Open Source, Open Access
biocontext7 is MIT-licensed and reproducible end-to-end.
- Website: biocontext7.com
- GitHub: Hordago-Labs/biocontext7
- npm (MCP server): @biocontext7/mcp
What’s Next
The registry expands weekly. Recent additions include batch-backfilled provenance manifests across all 2,061 skills, JOSS-ready documentation, and refreshed OG/branding assets. We’re grinding on:
- Expanding coverage in imaging and clinical genomics
- Tighter integration with Cursor and JetBrains
- Community skill submissions via PR
biocontext7 is a Hordago Labs project. We’re all in on AI for biology.