A malicious MCP server is silently stealing user emails
Koi Security says it's discovered the first malicious MCP server in the wild, exposing a risk to the entire ecosystem
Security researchers have issued a warning after discovering the first malicious Model Context Protocol (MCP) server in the wild.
MCP servers are widely used to allow AI agents to handle emails and run database queries, which means giving them access to all email traffic, according to Koi Security.
Postmark MCP Server is downloaded 1,500 times per week, and has been integrated into hundreds of developer workflows.
30% off Keeper Security's Business Starter and Business plans
Keeper Security is trusted and valued by thousands of businesses and millions of employees. Why not join them and protect your most important assets while taking advantage of this special offer?
However, since version 1.0.16 was released, Koi said it's been copying every single email - including invoices, internal memos, and confidential documents - to the developer's personal server.
"These MCP servers run with the same privileges as the AI assistants themselves - full email access, database connections, API permissions - yet they don't appear in any asset inventory, skip vendor risk assessments, and bypass every security control from DLP to email gateways," said Koi co-founder Idan Dardikman.
"By the time someone realizes their AI assistant has been quietly BCCing emails to an external server for months, the damage is already catastrophic."
There is a completely legitimate GitHub repo with the same name, officially maintained by Postmark, researchers noted. However, the attacker took the legitimate code from the repo, added his malicious BCC line, and published it to npm under the same name.
Sign up today and you will receive a free copy of our Future Focus 2026 report - the leading resource for IT decision-maker insight on priorities and investment areas in AI, security and more.
"This is the world’s first sighting of a real world malicious MCP server. The attack surface for endpoint supply chain attacks is slowly becoming the enterprise’s biggest attack surface," warned Dardikman.
"We're talking about 3,000 to 15,000 emails every day flowing straight to giftshop.club. And the truly messed up part? The developer didn't hack anything. Didn't exploit a zero-day. Didn't use some sophisticated attack vector,” he added/
“We literally handed him the keys, said 'here, run this code with full permissions', and let our AI assistants use it hundreds of times a day."
Malicious MCP server scrapped, but users still at risk
When Koi contacted the developer, Dardikman said he immediately deleted the package from npm. However, any organizations already using it will continue to be affected.
Users of version 1.0.16 should remove it immediately and rotate any credentials that may have been exposed, the company warned. They should also audit every MCP server they're using, making sure it's an official repository, reviewing the source code, and checking for changes in every update.
Koi said it reported the issue to npm, adding that the discovery raises broader concerns over MCP security.
"The postmark-mcp backdoor isn't just about one malicious developer or 1,500 weekly compromised installations. It's a warning shot about the MCP ecosystem itself," he said.
"We're handing god-mode permissions to tools built by people we don't know, can't verify, and have no reason to trust. These aren't just npm packages - they're direct pipelines into our most sensitive operations, automated by AI assistants that will use them thousands of times without question."
Make sure to follow ITPro on Google News to keep tabs on all our latest news, analysis, and reviews.
MORE FROM ITPRO
- The rise of GhostGPT – Why cybercriminals are turning to generative AI
- Microsoft quietly launched an AI agent that can detect and reverse engineer malware
- Generative AI attacks are accelerating at an alarming rate
Emma Woollacott is a freelance journalist writing for publications including the BBC, Private Eye, Forbes, Raconteur and specialist technology titles.
-
UK tech leaders are implementing AI faster than related governanceNews IBM research shows that chief AI officers are being given strategy roles, but not full control
-
11:11 Systems appoints new EVP in EMEA growth driveNews The experienced executive will oversee the provider’s operations and strategic execution across the region as it targets further growth
-
Can AI fight AI? Where the security gap still exists in cybersecurity, and how MSPs can help.Industry Insights Why AI security is failing and how MSPs can close the gap
-
After OpenAI-Hugging Face, how do IT leaders need to change the way they think about AI?In depth The OpenAI-Hugging Face incident and Anthropic admission soon after have opened up new conversations about controls around AI. How should IT leaders change their thinking about the technology?
-
Taking the myths out of Mythos - the role for the channel around AI and securityIndustry Insights Agentic security and vulnerability management must be a proactive priority rather than a reactive response to a problem already there
-
The OpenAI and Anthropic containment breaches are a bit spooky, but also quite sillyOpinion An AI leaving notes to future versions of itself is pure sci-fi; forgetting to lock down an environment is prosaic
-
Agent 009… the nine-second warningIndustry Insights AI Agents can go rogue. What is the channel’s role as guardians of recovery?
-
‘Every hour ransomware goes undetected drastically increases its potential blast radius’: Hackers are breaching networks and laying low for longer – and nearly half of firms don’t realize until data is stolenNews An ExtraHop survey found more intrusions are going undetected, leading to longer dwell times
-
Apple is speeding up software patching due to AI security concerns – here’s what you need to knowNews Apple is speeding up its software patching processes amid rising concerns that AI is helping hackers to spot and exploit flaws at a far quicker pace.
-
Microsoft joins competitors in handing over AI models for advanced testingNews US and UK government agencies will evaluate the firm's frontier models, along with those from Google and xAI
