For decades, enterprise software has revolved around dedicated management consoles. Whether investigating backup health, validating recovery readiness, reviewing compliance policies, or responding to operational issues, administrators have learned to navigate dashboards, reports, and alerts spread across multiple interfaces.
That model is beginning to change.AI assistants, copilots, and autonomous agents are rapidly becoming a new operational interface for enterprise software. Rather than switching between management consoles, administrators increasingly expect to ask questions in natural language and receive informed answers within the AI tools they already use.
Connecting AI to enterprise software is relatively straightforward. Giving AI the operational context needed to make informed decisions is much harder. Understanding backup health, recovery readiness, governance posture, or cyber resilience requires much more than access to individual APIs. It depends on understanding how backup activity, identities, policies, recovery operations, platform telemetry, and operational relationships fit together.
That's exactly the problem Druva MCP was designed to solve.
Druva MCP is an open and remote Model Context Protocol (MCP) server that securely connects enterprise AI assistants and agents to the trusted backup intelligence within the Druva Resilience Cloud. Rather than simply exposing APIs, Druva MCP makes backup intelligence available within the AI workflows where administrators increasingly choose to work.
The architecture behind that experience is equally important. Many MCP implementations expose hundreds of individual tools as platforms grow. Druva takes a different approach, combining a cloud-native skills architecture with a programmatic Code Mode model that reduces complexity, minimizes token consumption, and allows new capabilities to be delivered continuously from the cloud.
The Traditional AI Harness Setup
Let’s look at how a standard AI ecosystem operates under ordinary circumstances.
Typically, an organization deploys an AI harness or client—such as Claude Desktop, Cursor, or Microsoft Copilot. To give that client access to external data, it connects to an MCP server. In a traditional arrangement, this integration relies on two distinct layers:
- Tools: The raw programmatic endpoints or functions that the AI can execute.
- Skills/Plugins: The underlying organizational logic, scripts, or step-by-step instructions that teach the AI how and when to utilize those tools to solve a problem.
While this model works for small, isolated projects, deploying it across an enterprise infrastructure becomes considerably more difficult to manage.
Traditional MCP Complexities vs. The Druva Approach
Context Window Bloat & Costly Tokens
The standard approach to building an MCP server relies on exposing individual, narrowly scoped semantic tools for every single capability (e.g., creating separate, unique tools like get_backup_status or list_alerts). This strategy can break down at enterprise scale where there are 100s of APIs to work with. Exposing hundreds of semantic tools bloats the Large Language Model’s (LLM) context window, consuming additional context window capacity and increasing token usage just to explain the rigid API schemas before the user even finishes typing a prompt.
How Druva MCP Overcomes It
Druva addresses this challenge by adopting the programmatic "Code Mode" (Programmatic Tool Calling) pattern popularized by platforms such as Anthropic and Cloudflare. Powered by the Dru MetaGraph, Druva MCP compresses the entire interface into just five lean, multi-purpose programmatic primitives rather than hundreds of rigid functions. This significantly reduces the context footprint, lowering token consumption while simplifying how the language model interacts with the platform.
The Local Maintenance "Upgrade Tax"
In a typical setup, the AI harness requires customers to manage and install plugins or skills separately on their local environments. This means whenever the underlying platform updates an API, introduces a feature, or modifies a workflow, your engineering team has to manually update local configurations or download client-side patches across the entire environment to prevent the integration from breaking.
How Druva Overcomes It
Because Druva is a fully cloud-native SaaS platform, we expose the necessary operational skills alongside the tools within the exact same cloud-managed MCP server. Customers never have to install skills separately on their local harness. Your AI client simply queries our SaaS backend on demand. When Druva introduces new capabilities, they become immediately available—delivering continuous platform innovation without requiring local client updates.
Druva's five Multi-purpose Tools
By shifting the heavy lifting to our backend cloud, the AI assistant interacts using only five fundamental capabilities:
- list_skills: Lists all available data protection, governance, and resilience skills exposed by the server. These skills are fully managed by Druva so if there is a new capability that is released or signature of an existing API changes, we update the skills. Skills are centrally managed by Druva and updated alongside the platform, with appropriate human oversight.
- recommend_skill: The core intelligence hub. It automatically evaluates the user's natural language keywords or intent, searches a specialized backend vector database of skill headers, and points the AI to the exact right capability that it can ask Druva to execute.
- retrieve_skill: Dynamically pulls the precise skill resource required, which can include targeted execution instructions or scripts.
- run_read: Safely runs the code or script—whether generated dynamically by the client LLM or bundled with the skill—to provide the final structured output directly to the user. The script runs in a hardened completely isolated sandbox environment so that it never touches your actual backup data. This tool can only fetch information from the Druva environment.
- run_write: This tool can be used for performing operations and non-destructive modification tasks on your Druva environment. It runs the code or script generated dynamically by the client LLM or securely bundled as part of the skill.
Bringing Trusted Backup Intelligence into Everyday Operations
Druva MCP extends backup intelligence beyond the management console. Instead of requiring administrators to navigate dashboards for every investigation or report, trusted backup intelligence becomes accessible through any MCP-compatible AI client.
Unified Intelligence Where You Already Work
Administrators increasingly want to work from a single operational interface rather than moving between multiple management consoles. Whether troubleshooting an outage, investigating backup health, or reviewing recovery readiness, Druva MCP makes trusted backup intelligence available directly within AI assistants such as Microsoft Copilot, Claude Desktop, Cursor, or other MCP-compatible clients.
Natural Language Visibility & Reporting
Teams can quickly automate tedious compliance investigations, risk assessments, and health check summaries. For example, administrators can instantly issue commands like:
- "Provide a summary of the global backup success rate across all regions."
- "Create a daily health overview detailing total data backed up, success percentages, and active critical alerts."
- "Check the backup health status for all integrated SaaS applications."
- "Look at the audit logs and identify any unusual activities by admin [Name]."
- “Check if our backups are happening as prescribed by SOC2 requirements.”
Built for an Open AI Ecosystem
Because Druva MCP is built on the open Model Context Protocol, organizations aren't limited to a single AI assistant. The same trusted backup intelligence can be securely accessed from Microsoft Copilot, Claude Desktop, Cursor, VS Code, or future MCP-compatible clients without changing how Druva itself operates.
Operational Investigations
Druva MCP categorizes capabilities into prioritized enterprise skills—spanning endpoint health, SaaS apps, cyber resilience, and data governance. This structured visibility makes it significantly faster to pinpoint protection gaps, locate quarantined ransomware recovery snapshots, or audit legal hold statuses before risks escalate.
Enterprise-Grade Security and Governance Guardrails
Allowing external AI agents to interact with critical data infrastructure requires an ironclad containment strategy. Druva MCP extends our security-first architecture directly into agentic workflows without introducing security silos:
- Implicit RBAC via 3-Legged OAuth: The MCP server utilizes an OAuth 2.1 flow where the LLM acts strictly on behalf of the authenticated user. If an administrator lacks the privileges to view specific M365 backups in the traditional console, the underlying API natively blocks the AI agent from accessing it.
- Strict Code Sandboxing: Any script executed by the tool runs in a highly restricted, ephemeral isolate with strictly controlled outbound egress. This hardened sandbox is completely disconnected from the Druva VPC, ensuring an adversary cannot execute malicious lateral code.
- No Destructive Actions: To prevent unintended or accidental AI behaviors, destructive delete operations are explicitly blocked through Druva MCP, providing an additional layer of safety.
- Separation of functions: LLMs usually treat write operations as more critical and ask for an approval before performing those actions. That’s why we have exposed two distinct tools for fetching any data or performing non-destructive actions within the Druva environment.
The Future of AI-Driven Resilience
Druva MCP extends the Druva platform beyond the management console, making trusted backup intelligence securely available to AI assistants and agents where operational decisions are made. By combining a cloud-native Code Mode architecture with centrally managed operational skills, Druva delivers an integration that evolves alongside the platform without requiring customers to continually maintain local integrations.
Learn More
Druva MCP is one of the capabilities within Druva AI Resilience, helping organizations securely extend trusted backup intelligence into enterprise AI assistants and agents.
Learn more:
- Register to see Druva’s AI capabilities in action at the upcoming Virtual Summit
- Explore the Druva MCP solution.
- Take an interactive product tour
- Download the Druva MCP datasheet.