About biocontext7

Deep skill bundles for AI bioinformatics agents

Name: biocontext7
Author: Hordago Labs

biocontext7 is a deterministic skill compiler. It produces 2,000+ structured, hash-verified documentation bundles for bioinformatics tools — compiled from live upstream sources, served via MCP to AI coding agents. Not hand-crafted prompts. Not LLM summaries. Not hallucinated tool APIs.

The problem

Large language models hallucinate bioinformatics tool parameters, reference outdated Bioconductor APIs, suggest wrong conda channels, and fabricate function signatures. In computational biology, a single wrong parameter can silently corrupt an entire analysis pipeline.

Hallucinated function parameters that don't exist
Outdated API versions from training data cutoff
Wrong package managers and install commands
Fabricated citation DOIs and publication metadata

The solution

biocontext7 compiles 2,000+ deep skill bundles from live upstream documentation and serves them to AI agents via the Model Context Protocol (MCP). Every skill carries a SHA-256 source hash, upstream version, and generation timestamp so you can verify exactly what your agent is reading and when it was last built.

Compiled from live upstream tutorials and registry metadata
Four mandatory provenance fields on every skill
Deterministic compiler — no LLM in the build path
15 MCP tools, 19 federation servers

How it works

A four-layer pipeline that crawls authoritative registries, indexes tools offline, and serves verified documentation to AI assistants.

Crawl + Segment

Fetch upstream documentation and split into workflow-specific reference files

Render (Jinja2, no LLM)

Compile registry metadata and segmented references into structured SKILL.md bundles

Verify + Hash

Compute SHA-256 source_hash, write mandatory provenance frontmatter, gate through quality checks

Serve via MCP

Deliver skill bundles through 15 registered MCP tools to Claude Code, Cursor, VS Code, and JetBrains over stdio or HTTP

End result

Claude or Cursor receives verified, up-to-date tool documentation — correct parameters, current versions, and validated install commands — eliminating hallucinated APIs.

What sets BioContext7 apart

The catalog is tuned for bioinformatics workflows, registry provenance, and grounded installation details.

Deep skill bundles

2,063 skill bundles compiled from live upstream documentation — not hand-crafted prompts, not LLM summaries. Each bundle carries structured references, code examples, and install commands ready for agent consumption.

Verifiable provenance

Every compiled skill carries four mandatory frontmatter fields: upstream_version, generated_at, source_hash (SHA-256), and generator_version. Same inputs always produce the same hash.

Deterministic compiler

Crawler → Segmenter → Renderer → Quality Gate → Recompiler. No LLM in the build path. Skills can be regenerated byte-for-byte from a frozen crawl log for full reproducibility.

EDAM-typed I/O

Each skill bundle carries EDAM topics, operations, and format chains so agents can compose pipelines by input/output type compatibility, not just keyword match.

MCP-native

15 registered MCP tools backed by 19 specialist federation servers covering clinical, chemical, genomic, and literature databases. Works with Claude Code, Cursor, VS Code, and JetBrains.

Token-budgeted delivery

Skill bundles are rendered with per-file and aggregate token budgets so every response fits within the calling agent context window. No truncation surprises.

Built by Hordago Labs

Hordago Labs builds developer infrastructure for computational biology. biocontext7 is the flagship project: an open-source deep skill library that grounds AI coding assistants in hash-verified documentation — not trained guesses, not hallucinated tool APIs.

Open Source

2,063 Deep Skill Bundles

15 MCP Tools