Oktsec v0.6.0:
MCP Gateway, Security Hardening & Coordinated Stack Release

Coordinated release across the entire stack

Today we are shipping Oktsec v0.6.0, Aguara v0.4.0, and Aguara MCP v0.3.0 simultaneously. This is the first fully coordinated release across all three projects. Every binary, every library, every transport layer has been migrated from community MCP forks to the official modelcontextprotocol/go-sdk v1.4.0 under Linux Foundation governance. Zero community forks remain.

For Oktsec specifically, v0.6.0 is the largest release to date. It introduces a fourth operational mode -- the MCP Gateway -- alongside proxy, audit, and MCP server modes. It closes four distinct security vectors that were absent from the previous release. And it triples MCP client discovery from 6 to 17 supported clients.

This post covers everything that shipped. If you are evaluating Oktsec for production deployment, this is the release that makes the case.

MCP Gateway Mode

The headline feature. The new oktsec gateway command starts a Streamable HTTP server that fronts multiple backend MCP servers. Agents connect to the gateway. The gateway discovers tools from all connected backends, applies the full security pipeline, and proxies approved calls downstream.

Gateway capabilities at a glance:

This is a new operational mode alongside the existing proxy, audit, and MCP server modes. Each mode addresses a different deployment topology. The gateway is designed for environments where multiple backend MCP servers need centralized security enforcement with a single control plane.

Security Hardening

Four security vectors closed in this release. Each represents an attack surface that was either unprotected or absent from prior versions.

Combined effect: SSRF, credential leakage, replay attacks, and symlink-based file manipulation are now blocked at the infrastructure layer. These protections apply to all four operational modes -- proxy, audit, MCP server, and gateway.

Expanded MCP Client Discovery

Oktsec auto-discovers MCP client configuration files on the local system to enable one-command integration. In v0.6.0, supported clients expanded from 6 to 17:

This covers the major MCP-capable editors, IDEs, and CLI tools currently in production use. When new clients adopt the MCP protocol, Oktsec adds discovery support in the next release.

Dashboard & Webhook Enhancements

Redesigned UI

The dashboard login, 404 page, and root splash have been fully redesigned. The visual refresh aligns the monitoring interface with the operational gravity of the data it presents.

Agent metrics

• Risk scores -- Per-agent risk assessment based on triggered rule severity and frequency
• Traffic stats -- Request volume, blocked vs. allowed ratios, latency distributions
• Communication partners -- Which backends each agent communicates with and how often
• Top triggered rules -- The most frequently triggered detection rules per agent, ranked by category

Events page

• Agent filter -- Scope the event stream to a specific agent identity
• Time-range filter -- Narrow events to a specific window for forensic investigation

Overview dashboard

• Detection rate % -- Percentage of total requests that triggered at least one detection rule
• Unsigned % -- Percentage of requests arriving without Ed25519 identity signatures
• Average latency -- End-to-end pipeline processing time across all modes

Webhook improvements

Webhook payloads now include rule_name, category, and match fields for every triggered detection rule. Named webhook channels allow routing different event types to different destinations -- critical alerts to PagerDuty, audit events to Slack, compliance records to S3.

Official MCP SDK Migration

The entire codebase has migrated from mark3labs/mcp-go v0.44 to modelcontextprotocol/go-sdk v1.4.0. This is the official Go SDK maintained under the Linux Foundation's Model Context Protocol organization -- Tier 1 status with semver stability guarantees.

The migration was not a find-and-replace. All MCP server code, client code, and transport handling was rewritten. A new internal/mcputil helper package encapsulates common patterns for tool registration, request parsing, and response construction.

The governance shift matters. Community forks like mark3labs/mcp-go provided valuable early momentum, but they carry inherent risks: no guaranteed maintenance cadence, no formal security review process, no commitment to backward compatibility. The official SDK provides all three.

Zero community forks remain across the entire stack. Oktsec, Aguara, and Aguara MCP all run on the same official SDK. A single upstream fix propagates everywhere.

Coordinated Engine Upgrades

The Oktsec release ships alongside coordinated updates to the Aguara detection engine and monitoring platform:

• Aguara v0.4.0 -- 153 detection rules (+5 new), file and memory hardening with the same internal/safefile package, official SDK migration
• Aguara MCP v0.3.0 -- Same SDK migration to modelcontextprotocol/go-sdk v1.4.0, scan timeout enforcement, error message sanitization to prevent information leakage
• Aguara Watch -- 42,969 MCP skills tracked across 7 registries (+2 new registries), 99% Grade A security ratings across the monitored skill population

The Aguara-specific details are covered in the dedicated release post: Aguara v0.4.0: Coordinated Release.

By the Numbers

What's Next

The next release cycle focuses on production readiness and expanding into regulated environments:

• Binary releases via GoReleaser -- Pre-built binaries for Linux, macOS, and Windows. No Go toolchain required for deployment. Homebrew tap and APT repository on the same timeline.
• Gateway authentication -- API key and mTLS authentication for the gateway endpoint. The current header-based agent identity model works for local and development deployments; production environments need stronger authentication primitives at the transport layer.
• Targeting 200+ detection rules -- New categories across both engines covering supply chain attacks, tool schema manipulation, and multi-step exfiltration patterns. The detection engine is being restructured for dynamic rule loading without restart.
• Regulated environments. Payment systems, healthcare data flows, and compliance-driven verticals where every message between agents needs cryptographic attribution and a tamper-proof audit trail. The MCP Gateway's 9-step pipeline and SARIF output map directly to these requirements.

Every component ships under the same coordinated release cadence. When Aguara gains new detection rules, Oktsec gains them in the same release. When the SDK receives a security patch, the entire stack updates together.

The security pipeline is production-ready. If your AI agents are connecting to databases, APIs, payment systems, or file servers without a security layer in between, you have a problem. Every one of those connections is an attack surface. Oktsec sits in the middle and enforces security on every single call. Identity, policy, content scanning, audit. Nothing passes through without being checked.

If you're building agents for payment systems, healthcare data flows, or any regulated environment where every interaction needs identity, traceability, and a complete audit log: I want to hear about your use case.