AI for Business

MCP Servers Could Leak Enterprise Secrets - Here's What to Know

MCP servers - the middlemen that let AI agents tap into company tools and data - can quietly expose sensitive credentials, often before security teams even realize the server is running. The Model...

Share:
MCP Servers Could Leak Enterprise Secrets - Here's What to Know

MCP servers - the middlemen that let AI agents tap into company tools and data - can quietly expose sensitive credentials, often before security teams even realize the server is running. The Model Context Protocol, an open standard introduced by Anthropic, connects AI assistants to live systems like databases, files, and APIs. To do that, the server holds the keys: credentials, service account keys, API tokens, and other secrets.

The risk isn't hypothetical. Plaintext credentials often sit in local config files, sometimes pasted in during setup and forgotten. Those files can end up copied between machines or accidentally committed to Git. Credential sprawl is another issue - the same API keys and tokens get scattered across dev, staging, and production environments, rarely rotated, leaving long-lived secrets open to theft.

Prompt injection adds another layer: attackers can hide instructions inside documents or web pages an agent reads, tricking it into misusing tools or handing over secrets. Over-permissioning is common too, as developers grant broad scopes during building and forget to tighten them before shipping. And with anyone able to publish an MCP server, supply chain risks loom - CVE-2025-6514 showed how a malicious server could trigger remote code execution on machines running the mcp-remote OAuth proxy.

To secure these servers, experts recommend centralizing secrets in a managed store, using short-lived credentials with automatic rotation, enforcing least privilege, keeping a human in the loop for sensitive actions, encrypting secrets with a zero-knowledge model, logging all agent activity, and maintaining an inventory of every MCP server in use.

As AI agents become active identities across enterprise systems, a leaked secret doesn't just expose data - it grants attackers the ability to act on it. Organizations need to apply the same rigor to the MCP layer as they would to any other production system holding secrets.

Source: The Hackers News

Ready to Modernize Your Business?

Get your AI automation roadmap in minutes, not months.

Analyze Your Workflows →