generated: '2026-09-06'
method: searched
status: published
source: https://help.druva.com/en/articles/15654265-getting-started-with-the-druva-mcp-server
sources:
- https://help.druva.com/en/articles/15654265-getting-started-with-the-druva-mcp-server
- https://www.druva.com/use-cases/ai-resilience/ai/druva-mcp
- https://www.druva.com/blog/mcp-brings-backup-intelligence-ai
- https://mcp.druva.com/.well-known/oauth-authorization-server
- https://mcp.druva.com/.well-known/oauth-protected-resource/mcp
name: Druva MCP Server
description: >-
  Druva publishes a first-party hosted MCP server that connects MCP-compatible AI clients
  (Claude Desktop, Cursor, VS Code, Microsoft Copilot) to the Druva Resilience Cloud under
  the caller's own Druva RBAC. Rather than exposing one tool per REST operation, Druva ships
  a "Code Mode" skills architecture: five generic tools that discover, retrieve and execute
  cloud-delivered skills, so new capability is added server-side without a client update.
deployment:
  mode: remote
  endpoint: https://mcp.druva.com/mcp
  transport: streamable-http
  auth: oauth
  verified: probed
  probe:
    method: POST
    url: https://mcp.druva.com/mcp
    body: '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
    http_status: 401
    response: '{"error": "invalid_token", "error_description": "Authentication required"}'
    www_authenticate: >-
      Bearer error="invalid_token", error_description="Authentication required",
      resource_metadata="https://mcp.druva.com/.well-known/oauth-protected-resource/mcp"
    server: uvicorn
    date: '2026-09-06'
    note: >-
      The endpoint is live and speaks MCP over HTTP, but tools/list is OAuth-gated, so the
      live inputSchema for each tool could not be introspected anonymously. The tool names
      and semantics below are taken verbatim from Druva's own published documentation, NOT
      from an authenticated introspection. Schemas require an authenticated tools/list.
oauth:
  authorization_server: https://mcp.druva.com
  authorization_endpoint: https://login.druva.com/login
  token_endpoint: https://mcp.druva.com/token
  registration_endpoint: https://mcp.druva.com/register
  grant_types: [authorization_code, refresh_token]
  code_challenge_methods: [S256]
  token_endpoint_auth_methods: [none]
  scopes: [mcp:tools, mcp:resources]
  protected_resource: https://mcp.druva.com/mcp
  flow_documented_as: 3-legged OAuth 2.1 with implicit role-based access control
  discovery_documents:
  - well-known/druva-mcp-oauth-authorization-server.json
  - well-known/druva-mcp-oauth-protected-resource.json
client_config: |
  {
    "mcpServers": {
      "druva": {
        "type": "http",
        "url": "https://mcp.druva.com/mcp"
      }
    }
  }
tools:
- name: druva___list_skills
  description: List the Druva skills available to the authenticated caller.
  schema: auth-gated
- name: druva___recommend_skill
  description: Recommend the Druva skill that best matches a natural-language request.
  schema: auth-gated
- name: druva___retrieve_skill
  description: Retrieve the definition of a named Druva skill.
  schema: auth-gated
- name: druva___run_read_script
  description: Execute a read-only script against the Druva Resilience Cloud on the caller's behalf.
  schema: auth-gated
- name: druva___run_write_script
  description: >-
    Execute a configuration-modifying script. Druva documents that this cannot delete Druva
    Cloud configurations or backup data - destructive delete operations are blocked server-side.
  schema: auth-gated
tool_count: 5
skill_domains:
- endpoint health
- SaaS apps
- cyber resilience
- data governance
safety:
  destructive_operations: blocked
  note: >-
    Druva documents that the MCP server has permission to read and modify configuration
    settings but cannot delete Druva Cloud configurations or backup data, and that every
    interaction is constrained by the caller's existing Druva RBAC.