Your AI agents are running.
Nobody’s watching.
MCPSec is the security governance layer for Model Context Protocol — access control, audit logging, and anomaly detection for every tool call your AI agents make.
Open source core · No credit card · Enterprise waitlist open
MCP adoption is outpacing
security by 18 months
Engineering teams are connecting AI agents to databases, APIs, and internal systems via MCP. Security teams have no visibility into what those agents actually do.
No audit trail
Every tool call your AI agent makes is invisible. When a SOC2 or DORA auditor asks what your agent accessed last Tuesday — you have no answer.
No access control
Any connected model can call any tool, with any parameters. There is no policy layer between your AI agents and your most sensitive internal systems.
No anomaly detection
A compromised agent querying databases it has never touched, or making 100x its normal API calls — you won’t know until the damage is done.
No compliance story
DORA, NIS2, SOC2, and ISO 27001 all require documented ICT risk management. AI agents operating without governance are an unmanaged risk by definition.
A proxy layer that sees everything.
Changes nothing.
MCPSec sits between your AI agents and your MCP servers. Zero changes to existing infrastructure. Deploy in minutes.
Deploy the MCP Security Proxy OPEN SOURCE
A single sidecar container or DNS routing change puts MCPSec in front of your MCP servers. No refactoring. No downtime. Works with any MCP-compatible agent or framework.
Every tool call is intercepted and logged OPEN SOURCE
MCPSec parses every MCP tool call in real time — agent identity, tool name, parameters, response, session context, timestamp. Stored in a tamper-evident, cryptographically signed append-only log.
Access policies enforced at the edge OPEN SOURCE
Define which agents can call which tools under what conditions. Block, rate-limit, or require additional authentication — before the request ever reaches your MCP server.
Anomaly detection and SIEM integration ENTERPRISE
ML-based behavioural baselining flags deviations in real time. Native connectors for Splunk, Elastic, and Graylog push events directly into your existing SOC workflow.
One-click compliance reports ENTERPRISE
Pre-built audit packages for DORA, NIS2, SOC2, and ISO 27001. Generated from your actual agent activity log — not a template. Ready for your next audit.
Everything your security team
needs. Nothing they don’t.
Built by security engineers who have run enterprise SOCs. Designed to integrate with what you already have.
Real-time activity log
Every tool call logged with full context — agent, tool, parameters, response, user session, latency. Queryable by any dimension in seconds.
RBAC access control
Role-based policies defining which models can access which tools. Allowlists, denylists, rate limits, and conditional rules — all without touching your MCP servers.
Tamper-evident logging
Cryptographically chained log entries. Any modification to historical logs is immediately detectable. Audit-grade integrity for regulatory requirements.
Anomaly detection
Behavioural baseline per agent, per environment. Deviations — unusual tools accessed, abnormal call volumes, off-hours activity — trigger configurable alerts.
SIEM integration
Native connectors for Splunk, Elastic, and Graylog. AI agent activity shows up in your existing SOC dashboard — no new tools for your security team to learn.
Compliance reports
Pre-built audit packages for DORA, NIS2, SOC2, and ISO 27001 — generated from your real agent activity data. One click, audit-ready output.
Built for regulated environments
MCPSec is designed from the ground up to support the compliance frameworks that matter most to enterprise security teams in 2025 and beyond.
Start in minutes.
Trust what you can read.
Security teams don’t adopt black-box tooling. MCPSec’s core is Apache 2.0 licensed, auditable, and self-hostable forever.
Deploy the MCP Security Proxy alongside your existing infrastructure:
# Install MCPSec proxy npm install -g mcpsec # Start proxying your MCP server mcpsec proxy --target localhost:3000 --port 4000 # View live activity log mcpsec logs --follow
Apache 2.0 · Self-hosted · No telemetry by default · Full docs on GitHub
Be the first to know
when we ship.
We’re onboarding a small group of design partners — security teams already dealing with MCP governance in production. If that’s you, we want to talk.
Open source v0.1 shipping Q3 2026 · Enterprise early access Q4 2026 · No spam, ever