#mcp
-
A structural flaw in MCP lets malicious prompts spread between AI agents
Ars Technica warns that MCP, the protocol increasingly used for AI agents to talk to one another, carries trust gaps that let malicious prompts travel from one agent to the next.
-
Flaw in the Official MCP Python SDK Lets Malicious Servers Steal OAuth Credentials
A rogue MCP server could trick an app built on the official Python SDK into handing over its client secret, authorization code, and PKCE proof key. Fixes are in versions 1.30.0 and 2.2.0.
-
Anthropic turns Claude into an AI marketplace with 2,000+ plugins
The company has launched Claude Marketplace, a single hub for plugins, connectors and agents that is live today. Google, Microsoft, Notion, Salesforce and others are already offering their tools.
-
Critical Bifrost AI Gateway Flaw Allows Command Execution Without Credentials
A critical flaw in Bifrost, the open-source AI gateway that routes requests to over 20 LLM providers, lets an unauthenticated attacker run arbitrary commands on the gateway server with a single HTTP request.
-
Google Opens the Smart Home to Third-Party AI Agents via MCP
Google Home has gained support for the Model Context Protocol — now agents like Claude or Open Claw can analyze your home's data and control devices.
-
The Updated MCP Roadmap: Agents, Transport and Enterprise Security
The developers of the Model Context Protocol have published a new roadmap setting out how the protocol will evolve over the coming months. It is organized around five priority areas.
-
GhostSplice: How a Command Broken Into Pieces Makes an AI Agent Steal Secrets
Researchers at ASSET Research Group have shown that a malicious MCP server can coax out SSH keys and .env files even when the agent refuses a direct request to steal them — all it takes is to splinter the instruction into harmless fragments.
-
Patches for Veeam, Terraform MCP and Django: The Big Hole Is Cross-Tenant, at CVSS 10.0
HashiCorp, Veeam and the Django Software Foundation have closed 11 vulnerabilities. The most serious is a cross-tenant bug in Terraform MCP Server that lets someone reuse another user's token.
-
Stateless MCP: the protocol update that won back Simon Willison's interest
The July 28, 2026 revision of the Model Context Protocol did away with server-side sessions. Simon Willison explains why this simplifies development and makes agents safer — and shows off three tools of his own.
-
FakeGit: 7,600 Fake GitHub Repositories Are Now Baiting AI Agents Too
The FakeGit campaign distributes the SmartLoader downloader through bogus repositories masquerading as AI skills and MCP servers. A new technique called AgentBaiting deceives not people, but the AI agents themselves.