Skip to content

DEV Community

← All Trends

Model Context Protocol (MCP) Security

13 posts in this trend in the last 7 days • Active about 3 hours ago

Jeff

Oct 3

MCP Security in Practice: Prompt Injection, Least Privilege, and Audit Logs

#mcp #security #ai #llm

5 min read

Neeraj Ram

Oct 8

MCP Servers Are a New Attack Surface, and Most Teams Are Not Looking

#ai #security #mcp #llm

5 min read

Kiell Tampubolon

Oct 4

Threat Modeling the Model Context Protocol: Securing Agentic Tools with mcpscan

#security #ai #mcp #python

5 min read

Oct 6

Prompt injection is a data plane problem

#security #aiagents #mcp #llm

7 min read

Oct 2

We Trusted the Tool Descriptions. That Was the Bug.

#mcp #ai #frontend #security

11 min read

Oct 4

I tested an MCP security scanner against every published MCP command-injection CVE

#security #mcp #ai #opensource

3 min read

TechSavant Security Lab

Oct 8

5 MCP Security Tests to Run Before Your AI Agent Gets Tool Access

#security #ai #llm #mcp

3 min read

Royal Simpson Pinto

Oct 7

Building a Minimal MCP Server with Scoped Tools and Per-Client Boundaries

#mcp #security #typescript #tutorial

5 min read

Harsh Shah

Oct 4

Model Context Protocol (MCP): The New Standard for AI Tool Integration

#ai #mcp #security #architecture

10 min read

mech.app

Oct 8

MCP Agent-to-Agent Vulnerabilities: What Google's Flaw Reveals About Protocol-Level Security Boundaries

#agents #cybersecurity #mcp #security

5 min read

AgentAvow

Oct 2

We started running every MCP server we grade. Here's what 20 popular ones actually did.

#ai #security #mcp #agents

4 min read

Oct 6

Indirect Prompt Injection Through Tool Descriptions and Tool Output: How Untrusted Metadata Hijacks Agents

#ai #agents #security #llm

7 min read

JUICE

Oct 7

Why Your MCP Filesystem Server Needs a Guardrail (and How to Build One in Rust)

#rust #mcp #security #tooling

4 min read

Trend Analysis

As developers rapidly adopt Anthropic's Model Context Protocol to connect AI agents to local files, databases, and internal tools, a critical new attack surface has emerged. Articles explore the risks of prompt injection, untrusted tool descriptions, and the urgent need for threat modeling, least privilege enforcement, and audit logs.

Core Questions Under Debate:

How can we prevent indirect prompt injections via untrusted tool descriptions?
What strategies ensure least privilege and proper access control for MCP servers?
How do we effectively threat model and audit AI agent tool interactions?

First detected: October 07, 2026

Last updated: about 3 hours ago

Algorithm: Vector Leader Clustering (threshold: 82%)

About this trend

This trend groups posts from the last 7 days that discuss similar concepts, technologies, or ideas.