mTrust provides a managed AI firewall for the Model Context Protocol (MCP), protecting both MCP servers from untrusted agents and agents from malicious servers. The platform intercepts every request between AI agents and tools, calculating real-time trust scores, scanning for prompt injections, and learning normal behavioral patterns. It deploys in minutes and functions as a security layer for the rapidly growing MCP ecosystem.
Funding
Funding not disclosed
Founders
Product
Problem
MCP (Model Context Protocol) is becoming the standard way AI agents interact with software, yet it has no built-in security layer. When an agent connects to an MCP server, there is no identity check, trust score, behavioral analysis, or audit trail — the server simply executes whatever the agent requests. A compromised agent can exfiltrate data, execute unauthorized commands, and pivot through connected systems with 80-90% autonomy and zero human oversight.
Solution
mTrust is a managed AI firewall that sits between AI agents and tools, intercepting every request. It identifies the agent, calculates a real-time trust score based on behavioral patterns, enforces security policies, and learns what "normal" looks like for each environment. The platform protects in both directions: it shields MCP servers from untrusted agents and safeguards agents from malicious MCP servers. By combining trust scoring, injection scanning, and a global threat feed, mTrust catches attacks that rule-based systems miss. The solution deploys in minutes and provides continuous monitoring and adaptive policy enforcement as agent behavior evolves.
Target Audience
Primary customers are engineering and security teams at organizations deploying MCP servers or building AI agents, including enterprises using ChatGPT, Claude, or other AI assistants to access internal tools and data.
Features
- Real-time trust scoring for AI agents based on behavioral analysis and historical patterns
- Prompt injection scanning that detects and blocks malicious instructions before execution
- Global threat feed aggregating known malicious agents, servers, and attack patterns across the MCP ecosystem
- Bidirectional protection covering both MCP server-to-agent and agent-to-server traffic
- Behavioral learning engine that establishes baselines and flags anomalies that rule-based systems miss
- Audit trail and logging for every agent request, enabling post-incident investigation and compliance reporting