Provides read-only access to Velero backup and schedule resources in Kubernetes clusters, enabling inspection of backups, schedules, and generation of Velero Backup YAML manifests through the Kubernetes API.
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:
π Inspect Velero backups
π Inspect Velero schedules
π Generate Velero Backup YAML manifests
π§© Access Velero data as MCP resources
π 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.
β 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
π Features
π§ 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.
π¦ MCP Resource Endpoints
Resource | Description |
| All backups in default namespace |
| All schedules in default namespace |
These allow LLMs to explore Velero state without calling tools.
π 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:
π 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:
π₯ Installation
For development:
βοΈ Configuration
Environment Variables
Variable | Description | Default |
| Path to kubeconfig | auto |
| Velero namespace | velero |
K8s auth order:
In-cluster ServiceAccount
$KUBECONFIG~/.kube/config
βΆοΈ Running the Server
Start the server in stdio mode (required by MCP):
π§© Example MCP Client Configuration
ChatGPT MCP configuration
Claude Desktop
π§ͺ Example Usage (AI Agent)
List failed backups
βCall
list_velero_backupsand filter phase = Failed.β
Inspect a backup
βUse
get_velero_backupforprod-fulland tell me included namespaces.β
Generate a manifest
βGenerate a Velero backup YAML including namespaces
db,logging, TTL=168h.β
Produces:
π Development
Run linters:
Type-check:
Run tests:
π€ Contributing
We welcome contributions of all kinds:
Add Velero Restore / VolumeSnapshot support
Improve error handling
Add more MCP resources
Add a Helm chart
Add logging or metrics
Improve documentation
Read the full CONTRIBUTING.md.
π License
This project is licensed under the MIT License, allowing:
Commercial use
Private use
Modification
Distribution
πΊ Roadmap
Planned improvements:
Add support for Restore CRDs
Snapshot objects
Restore impact analysis
Write-enabled mode (behind strict flags)
Publish to PyPI
Helm chart for cluster deployment
If you'd like to help shape the direction, please open an issue!