Bizard Skills
AI-Ready Visualization Skills for Biomedical Research
What are Bizard Skills? 🧠
Bizard Skills is an AI skill package that turns your AI assistant into a biomedical visualization expert. When you load this skill, your AI can:
- Recommend the best visualization type for your data and research question
- Generate reproducible R / Python / Julia plotting code adapted from Bizard tutorials
- Search across 1,000+ curated figure examples using the included gallery data
- Link you to the full tutorial for detailed customization
The skill package works with any AI tool that supports custom instructions or knowledge files:
- ChatGPT / GPT-4: Load as custom instructions + knowledge file
- Claude: Upload as project files
-
GitHub Copilot Chat: Reference with
#file:SKILL.md - Local LLMs (Ollama, LM Studio): Load as system prompt context
If the Bizard atlas — including this skill package and the MCP server described below — helps your research, please cite our paper:
Li, K., Zheng, H., Huang, K., Chai, Y., Peng, Y., Wang, C., … & Wang, S. (2026). Bizard: A Community-Driven Platform for Accelerating and Enhancing Biomedical Data Visualization. iMetaMed, e70038. https://doi.org/10.1002/imm3.70038
@article{li2026bizard,
title = {Bizard: A Community-Driven Platform for Accelerating and Enhancing Biomedical Data Visualization},
author = {Li, Kexin and Zheng, Hu and Huang, Kexin and others},
journal = {iMetaMed},
year = {2026},
pages = {e70038},
doi = {10.1002/imm3.70038}
}Quick start: Download the ZIP below → Load SKILL.md as your AI’s instructions → Upload gallery_data.csv as a knowledge file → Ask: “Help me create a volcano plot for my differential expression results”
Download the Skill Package
The ZIP contains:
-
SKILL.md— AI skill instructions: teaches your AI assistant how to use Bizard for visualization recommendations and code generation -
gallery_data.csv— 1,000+ visualization examples with image URLs, tutorial links, descriptions, and categories -
gallery_data_zh.csv— Chinese version of the gallery data -
README.md— Installation guide for different AI tools
Bizard MCP Server
For AI agents (Claude Code, Cursor, Cline, …) that can call tools directly, Bizard also ships a Model Context Protocol (MCP) server: @openbiox/bizard-mcp.
Unlike the static skill package above, the MCP server serves the live chart lifecycle — bilingual search across 1,000+ chart tutorials, runnable code with dependencies and data URLs, structured adaptation guidance, citation metadata, and template-checked contribution — from a snapshot generated by CI from the same sources as this website and SKILL.md, so all three never drift.
The skill and the MCP server work together. They are generated by CI from the same sources, and the MCP prompts (recommend, adapt, troubleshoot, contribute) implement the same workflow the skill teaches — so they never contradict each other. The skill plays the playbook role (chart-selection logic and gallery data, usable in any AI tool); the MCP server adds live, exact lookups (search hits with tutorial URLs, runnable code, adaptation checklists, citations, contribution submission). An agent that supports both — e.g., Claude Code with the skill installed and the server registered — follows the skill’s workflow while grounding every answer in MCP tool results.
No local install needed — any MCP-capable agent (Claude Code, Cursor, Cline, Windsurf, VS Code Copilot Chat, …) can connect to the public hosted instance deployed by the Bizard team. Most clients accept a JSON server entry:
{
"mcpServers": {
"bizard": { "type": "http", "url": "http://biotree.top:38123/bizard-mcp/mcp" }
}
}In Claude Code, one line is enough:
claude mcp add --transport http bizard http://biotree.top:38123/bizard-mcp/mcpPrefer a fully local setup? Install the package instead:
claude mcp add bizard -- npx @openbiox/bizard-mcpOr run it with any MCP client over stdio (npx @openbiox/bizard-mcp) or your own Streamable HTTP instance (npx @openbiox/bizard-mcp --http). See the MCP server docs below for the full tool list, configuration, and privacy notes.
MCP Server Details
The full documentation for @openbiox/bizard-mcp is maintained in the repository and reproduced here for reference.
Tools
| Tool | Purpose |
|---|---|
bizard_search |
Bilingual weighted search over the chart gallery; adjacent-category suggestions on zero hits |
bizard_list_categories |
All 58 chart categories (Level 2 types) with per-category chart counts (bilingual names where available) |
bizard_get_tutorial |
Chart metadata: packages, example-data URLs, available languages, tutorial link |
bizard_get_code |
Runnable tutorial code with package list, example-data URLs, and license notice |
bizard_adapt_guidance |
Data-shape checks, dependency prep, parameter mapping, failure modes, alternatives |
bizard_cite |
Bizard paper + tutorial citation (bibtex/text/json) with CC-BY-NC-4.0 license notice |
bizard_contribute_check |
Validate a draft against the contribution template; preview points |
bizard_contribute_submit |
Submit as issue/PR via user credentials, or degrade to draft + deep-link |
Prompts: recommend, adapt, troubleshoot, contribute — chartId arguments autocomplete via completion/complete. Resources: bizard://catalog, bizard://tutorial/{chartId} (enumerable via resources/list), bizard://llms-txt.
If charts or code from Bizard end up in your research, please cite the paper above — the bizard_cite tool returns ready-to-paste BibTeX/APA-style citations.
Deployment
One package, two transports:
-
Local (stdio, default):
npx @openbiox/bizard-mcp. Full capability; contribution runs at level 1/2 when local credentials exist. -
Hosted (Streamable HTTP):
npx @openbiox/bizard-mcp --http. Port viaBIZARD_HTTP_PORT(default 3000). Anonymous, stateless. Contribution runs at level 3 (no credentials on the server). -
Public hosted instance:
claude mcp add --transport http bizard http://biotree.top:38123/bizard-mcp/mcp— deployed by the Bizard team; no install required and may be faster to reach from mainland China. Same anonymous/stateless/rate-limited behavior, contribution at level 3.
Contribution credential degradation
bizard_contribute_submit probes the environment on every call and states which level it used — never silently failing:
| Level | Condition | Behavior |
|---|---|---|
| 1 |
gh CLI installed and logged in |
Creates the issue as your GitHub user (PR submission is planned for a future release) |
| 2 | No gh, but GITHUB_TOKEN set |
GitHub API submission, limited by token scope |
| 3 | Neither (typical for hosted) | Returns formatted draft + pre-filled issues/new?title=&body= deep-link |
The server itself holds zero credentials. User free text is truncated (single field ≤ 4000 chars) before being assembled into issue bodies.
Privacy
Usage feedback is local-only, anonymous, and never networked:
- Appended to
~/.cache/bizard-mcp/usage-log.jsonlas{ ts, event, chartId, language, queryHash }—queryHashis the first 8 hex chars of the query’s SHA-256 (irreversible). - Set
BIZARD_USAGE_LOG=offto disable entirely; set it to a path to relocate the log. - No server-side telemetry, ever.
Source code and development instructions: openbiox/Bizard · mcp/.
Regenerating Skills
SKILL.md is automatically regenerated when tutorials are updated on the main branch. To regenerate manually:
# LLM mode (requires API key in environment)
python .github/scripts/generate_skills.py --verbose
# Offline mode (rule-based extraction)
python .github/scripts/generate_skills.py --offline --verboseSee the Skill Generation Specification for the full technical specification.