MCP 服务器

wDVC MCP

Model Context Protocol server for wDVC that provides tools for architecting DVC pipelines, generating Docker worker commands, searching patterns, and scaffolding wDVC projects.

README

wDVC MCP

wDVC MCP — Model Context Protocol server for wDVC (DVC Data Pipeline Management). Provides tools for architecting DVC pipelines, generating Docker worker commands for data download, and scaffolding wDVC projects.

Features

Feature	Description
Architect Blueprints	Complete reference for DVC pipeline stages, worker queue, Gradio API, file downloader
Docker Worker Command	Generate exact `docker run` command for data download with configurable resources
API Usage	Gradio web UI examples for queue submission and status monitoring
Pattern Catalog	Searchable patterns from official/community repos with local fallbacks
Project Scaffolding	Generate complete wDVC project structure with templates
Type Safety	Fully typed, mypy clean

Installation

pip install wdvc-mcp

Or for development:

git clone https://github.com/wisrovi/wDVC-mcp.git
cd wDVC-mcp
pip install -e ".[dev]"
pre-commit install

Quick Start

Run the MCP Server

# Stdio transport (for Claude Desktop, etc.)
wdvc-mcp

# SSE transport (for HTTP clients)
wdvc-mcp --transport sse --port 8000

Available Tools

Tool	Description
`get_wdvc_architect_blueprints()`	Complete reference for all wDVC patterns
`get_wdvc_worker_command()`	Generate Docker run command for data download
`get_wdvc_api_usage()`	Gradio web UI usage examples
`search_wdvc_patterns(query)`	Search pattern catalog

Example: Get Docker Worker Command

from wdvc_mcp.server import get_wdvc_worker_command

# Default command
cmd = get_wdvc_worker_command()
print(cmd)

Output:

mkdir -p ./projects

docker run -it --rm \
  --name worker \
  --hostname wDVC \
  --shm-size=16g \
  --cpus="4.0" \
  --memory="4g" \
  -e IP_HOST=192.168.1.84 \
  -e REDIS_HOST=192.168.10.108 \
  -v ./projects:/app/projects \
  -w /app \
  wisrovi/dataset-ia:worker-v1 \
  zsh

Customize the Command

cmd = get_wdvc_worker_command(
    ip_host="10.0.0.1",
    redis_host="10.0.0.2",
    projects_path="/data/my_projects",
    image="myorg/dataset-ia:latest",
    cpus="8.0",
    memory="16g",
    shm_size="32g",
)

Project Scaffolding

Generate a complete wDVC project structure:

from wdvc_mcp.templates import TemplateGenerator

bp = TemplateGenerator.get_files_blueprint("standard", "my_pipeline")
# bp contains: main.py, config/settings.py, worker/worker.py, api/api.py,
# dvc.yaml, Dockerfile.worker, docker-compose.worker.yaml, run_worker.sh, etc.

Scaffold Types

Type	Description	Folders
`standard`	Full worker + API + config	config, worker, api, scripts, tests, .wdvc
`worker_service`	Worker only	config, worker, scripts, tests, .wdvc
`api_service`	API only	config, api, tests, .wdvc
`full_pipeline`	Everything + CI/CD	config, worker, api, scripts, pipeline, tests, examples, .wdvc, .github/workflows

Architecture

��─────────────────────────────────────────────────────────────��
│                      wDVC Architecture                       │
├─────────────────────────────────────────────────────────────��
│                                                              │
│  ��──────────────��    ��──────────────��    ��──────────────��  │
│  │  Gradio UI   │    │  Python SDK  │    │  MCP Tools   │  │
│  │  (api.py)    │    │  (worker.py) │    │  (server.py) │  │
│  └──────��───────��    └──────��───────��    └──────��───────��  │
│         │                   │                   │           │
│         └───────────────────��───────────────────��           │
│                             ��                                │
│                  ��─────────────────────��                    │
│                  │   Redis (wredis)    │                    │
│                  │  Queue + Hash +     │                    │
│                  │  SortedSet          │                    │
│                  └──────────��──────────��                    │
│                             │                                │
│         ��───────────────────��───────────────────��           │
│         ��                   ��                   ��           │
│  ��─────────────��    ��─────────────��    ��─────────────��    │
│  │ Docker      │    │ DVC Pipeline│    │ S3 Remote   │    │
│  │ Worker      │    │ (dvc.yaml)  │    │ (DVC push)  │    │
│  │ (container) │    │             │    │             │    │
│  └─────────────��    └─────────────��    └─────────────��    │
│                                                              │
��─────────────────────────────────────────────────────────────��

Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
make test

# Run with coverage
make test-cov

# Lint & format
make lint
make format

# Type check
make typecheck

# Build package
make build

# Publish to PyPI
make publish

Configuration

Environment Variables

Variable	Default	Description
`REDIS_HOST`	`localhost`	Redis server hostname
`REDIS_PORT`	`6379`	Redis server port
`REDIS_DB`	`0`	Redis database number
`REDIS_PASSWORD`	`None`	Redis password
`IP_HOST`	Auto-detected	Worker IP for registration