[Plugin Submission] Request to add kimi-protein-design to curated plugins

[devxia]

Plugin Submission

I would like to submit kimi-protein-design for consideration as a curated plugin in the Kimi Code ecosystem.

Repository

https://github.com/devxia/kimi-protein-design

One-line Description

End-to-end protein design pipeline — from backbone generation to structure validation, entirely through natural language conversation.

What It Does

kimi-protein-design is a Kimi Code plugin that integrates three state-of-the-art open-source protein design tools into a unified, conversational workflow:

Stage	Tool	Purpose
Stage 0	PDBFixer	Mandatory PDB preprocessing (repairs non-standard residues, removes heterogens, adds missing heavy atoms)
Stage 1	RFdiffusion	Protein backbone generation (monomers, binders, motif scaffolding, symmetric oligomers)
Stage 2	ProteinMPNN	Amino acid sequence design on fixed backbones
Stage 3	AlphaFold3	Structure prediction and confidence validation (pLDDT, ipTM, pTM)
Stage 4	Filtering	Automated quality filtering and ranking

Architecture Highlights

• MCP Server (stdio JSON-RPC 2.0) exposes 16 tools via mcp_server/server.py
• Async Job Manager with thread-pool execution, progress tracking, timeouts, and cleanup
• File-system + Historical ETA Progress Tracking — monitors output files and learns from past runtimes to give accurate progress estimates
• Conda Environment Support — each tool can run in its own isolated conda env via conda run
• Interactive Tool Configuration — if tools are missing, the plugin guides users to install and configure paths, all within the Kimi conversation
• AlphaFold3 Genetic Database Detection — validates public_databases presence and guides setup
• 7 Workflow Skills covering every stage plus end-to-end pipeline orchestration
• Recommended Hooks for context injection, GPU safety checks, and desktop notifications
• Bilingual Documentation (English + Chinese)

Target Kimi Code Version

>= 0.6.0 (uses Plugin GitHub install, Scheduled Tasks via CronCreate, no default step limit)

Installation

/plugins install https://github.com/devxia/kimi-protein-design
/new

Example Usage

User: Design a binder targeting PD-L1
→ Stage 0: PDBFixer preprocesses target.pdb
→ Stage 1: RFdiffusion generates 10 binder backbones
→ Stage 2: ProteinMPNN designs 8 sequences per backbone
→ Stage 3: AlphaFold3 validates top designs
→ Stage 4: Filter by ipTM > 0.8 and pLDDT > 80

Users can adjust output counts through natural language:

• "Generate 50 backbones" → num_designs = 50
• "16 sequences each" → num_seq_per_target = 16

Why It Should Be Curated

1. Scientific computing is a major use case for AI agents — protein design is a perfect fit for Kimi Code's long-running, multi-step workflow capabilities
2. Zero vendor lock-in — all underlying tools (RFdiffusion, ProteinMPNN, AlphaFold3, PDBFixer) are open-source and run locally
3. Production-ready async architecture — submit_job + query_job + CronCreate for batch validation of hundreds of designs without blocking the session
4. Active maintenance — well-documented, type-annotated, with robust error handling and user onboarding flows
5. Fills a gap — there is no other Kimi Code plugin for computational biology / structural bioinformatics

Checklist

• kimi.plugin.json manifest is valid
• README with installation and usage instructions
• English and Chinese documentation
• MCP server with tools/list and tools/call support
• Skills for workflow guidance
• Async job management with progress tracking
• Tested on Ubuntu 18.04 with Python 3.10+