MCP 服务器

Velero MCP Server

Provides read-only access to Velero backup and schedule resources in Kubernetes clusters, enabling AI agents to inspect backups, schedules, and generate Velero YAML manifests safely without write permissions.

README

Velero MCP Server

The Velero MCP Server is an open-source Model Context Protocol (MCP) server that exposes read-only, safe, structured access to Velero backup and schedule resources running inside any Kubernetes cluster.

It allows AI agents (ChatGPT, Claude, Cursor, GitHub Copilot, etc.) to:

🔍 Inspect Velero backups
🔄 Inspect Velero schedules
📄 Generate Velero Backup YAML manifests
🧩 Access Velero data as MCP resources
🔐 Safely interact with your cluster in read-only mode

This project helps platform engineers automate workflows using AI while ensuring zero-risk, low-privilege, and read-only access to critical cluster configuration.

⭐ Why This Project Exists

Velero is commonly used for:

Kubernetes namespace & cluster backups
Disaster recovery
Cluster migrations
Persistent volume snapshot management

But until now, no MCP server existed to expose Velero CRDs to LLM-based tools in a safe, structured way.

This project provides:

A consistent API for querying Velero
Strong typed models
Complete read-only safety
Guaranteed LLM-friendly output
Ready integration with GitOps

🚀 Features

🔧 MCP Tools

`list_velero_backups(namespace?: str)`

Returns a list of Velero Backup CRs.

`get_velero_backup(name: str, namespace?: str)`

Returns a detailed structured backup object.

`list_velero_schedules(namespace?: str)`

Lists Velero Schedule CRs including cron, paused state, and last backup.

`generate_velero_backup_yaml(...)`

Generates read-only YAML for a Velero Backup.

📦 MCP Resource Endpoints

Resource	Description
`velero://backups`	All backups in default namespace
`velero://schedules`	All schedules in default namespace

These allow LLMs to explore Velero state without calling tools.

🏗 Architecture

Python 3.10+
MCP (official Model Context Protocol SDK)
Kubernetes Python Client
Pydantic models
Safe, read-only design
No kubectl, no exec, no side effects

Flow:

MCP Client → Velero MCP Server → Kubernetes API → Velero CRDs

🔐 Security Model

Designed to be 100% safe

The server never:

Creates backups
Runs restores
Deletes backup objects
Writes anything to Kubernetes

Only reads CRDs via the Kubernetes API.

RBAC Required:

get, list on:
- backups.velero.io
- schedules.velero.io

📥 Installation

git clone https://github.com/YOUR-ORG/velero-mcp-server.git
cd velero-mcp-server

python -m venv .venv
source .venv/bin/activate    # Windows: .venv\Scripts\activate

pip install -r requirements.txt
pip install .

For development:

pip install ".[dev]"

⚙️ Configuration

Environment Variables

Variable	Description	Default
`KUBECONFIG`	Path to kubeconfig	auto
`VELERO_NAMESPACE`	Velero namespace	velero

K8s auth order:

In-cluster ServiceAccount
$KUBECONFIG
~/.kube/config

▶️ Running the Server

Start the server in stdio mode (required by MCP):

python -m velero_mcp_server.server

🧩 Example MCP Client Configuration

ChatGPT MCP configuration

{
  "mcpServers": {
    "velero-mcp": {
      "command": "python",
      "args": ["-m", "velero_mcp_server.server"],
      "env": {
        "KUBECONFIG": "/path/to/kubeconfig",
        "VELERO_NAMESPACE": "velero"
      }
    }
  }
}

Claude Desktop

"mcpServers": {
  "velero-mcp": {
    "command": "python",
    "args": ["-m", "velero_mcp_server.server"],
    "env": {
      "KUBECONFIG": "/path/to/kubeconfig",
      "VELERO_NAMESPACE": "velero"
    }
  }
}

🧪 Example Usage (AI Agent)

List failed backups

“Call list_velero_backups and filter phase = Failed.”

Inspect a backup

“Use get_velero_backup for prod-full and tell me included namespaces.”

Generate a manifest

“Generate a Velero backup YAML including namespaces db, logging, TTL=168h.”

Produces:

apiVersion: velero.io/v1
kind: Backup
metadata:
  name: prod-backup
  namespace: velero
spec:
  includedNamespaces:
    - db
    - logging
  ttl: 168h