Aegis Gateway 2.0

A configuration-driven governance layer for MCP calls. It gives agents one controlled route to real systems, with typed resources, approvals, destructive-action receipts, and locally verifiable audit evidence.

Start the Hello World Open the user handbook

Aegis own-host and Apache Knox integration choices share a fixed governance core before configured backend sources.
Start from a working profile. Aegis ships named configurations, IDE tutorials, and retained own-host and Apache Knox runs for the systems below. Read the validation report, claims matrix, and adopter catalog to see which target version each profile exercised.

Start with your task

Hello World

Put Aegis into your IDE and Apache project development cycle: build, start a profile, point MCP at Aegis, send one prompt.

User handbook

A concise route map for application teams, operators, security reviewers, and plugin authors.

Security and threat model

Trust zones, STRIDE abuse cases TM01 to TM10, OWASP MCP Top 10 handling, S1 to S31 promises, and dated control-suite records with their attestation status.

Threat modeling and honest coverage

The complete control plane, the controls enforced today, and a direct list of the open findings, the components Aegis does not own, and the checks that need a cluster or runtime.

MCP CVE containment demo

Three real MCP server CVE classes exploited directly, then denied or contained through Aegis, with before-and-after evidence from a real run.

Deploy on AWS and your AI IDE

Run the gateway image on EKS, ECS or EC2, then connect Kiro, Cursor, VS Code or Claude Desktop as a governed MCP server.

Responsible AI

Where Aegis contributes: human oversight, accountability, privacy and least privilege at the MCP action boundary, mapped to NIST AI RMF and the EU AI Act.

All tutorials

Deep-dive walkthroughs and one IDE-ready tutorial per verified Apache project, each from a retained real run.

Module reference

What each core root-reactor module does, its principal configuration keys, and the security controls it enforces.

Adopt Aegis

Connect an MCP server, declare REST tools, or choose a typed source without changing the governed project.

Verify evidence

See dated regression records, test boundaries, and real-engine evidence classifications.

Developer guide

Build, configure, connect an IDE, operate approvals and audit, and write sources, hooks, state and secret providers while preserving the engine’s fixed governance boundary.

Learn from working integrations

Start with the system you already operate. These guides connect a practical configuration, a governed MCP surface, and the evidence that describes what was exercised.

Choose a configuration

Find reference profiles for an existing MCP server, REST API, A2A agent, JDBC source, or Apache system. Each profile states its source type and evidence boundary.

Browse tutorials

Follow guided own-host and Knox paths for Kafka, Flink, Iceberg, Spark, ZooKeeper, and other verified project integrations.

Connect your IDE

Open the repository tutorial folder for ready-made VS Code, Kiro, and Cursor MCP settings. Start the matching gateway first and supply fresh local credentials.

Inspect retained examples

Review the exact configuration, topology, engine settings, audit records, and result logs from retained integration runs. Evidence status remains explicit for every profile.

How a destructive action becomes executable

Evaluate, preview, independently approve, commit audit intent and execute. Any denial prevents the backend call.

Delivery evidence at a glance

SurfaceEvidence statusWhere to go next
Core governanceDefault build on 2026-10-03: 364 tests in a 20-project reactor, 0 failures, 0 errors, 0 skipped, and a clean Apache RAT check.Validation report
Adapter catalog25 deployable adapter configurations and 116 tools, parsed and governed in the default build.Repository adopter catalog
Real-engine runsTwenty-eight profiles have own-host and Knox runs, each with an IDE-ready tutorial. Four more ship as configuration for the next product-stack run: Atlas, Cassandra, Impala, and Paimon.Adopter status
Security and threat modelThreat model (STRIDE, TM01 to TM10), security properties S1 to S31, and a focused control suite of 227 tests with 0 failures.Security guide
Security hardening and build gates (2.0)Hardened build on 2026-11-07: 408 tests across the eight core modules, 0 failures, with SpotBugs and the FindSecBugs plugin, JaCoCo coverage, and the Apache RAT license check all green. CodeQL and an OSV dependency scan are added as CI workflows.Threat modeling
MCP CVE containmentThree real MCP server CVE classes (filesystem path traversal, command injection, arbitrary file access) exploited directly, then denied or contained through Aegis, with before-and-after evidence from a real run.CVE containment
Added 2026-10-03Per-caller backend credentials by RFC 8693 token exchange (Keycloak 26.8.0), A2A agents as governed tools (a2a-sdk 1.2.1 through both front doors), an OpenAPI-to-draft command, three reference hook services, and a container image and Helm chart.Runs added on 2026-10-03
Knox extensionKnox 3.0.0 API-level tests, and a 2026-10-05 run of the current bundle inside a stock Knox 3.0.0 gateway against real ZooKeeper. Knox 3.1 needs a Jakarta build. An operator must validate their Knox topology and plugin set.Knox extension guide
How to read a profile: each adopter page names the target version that was exercised and links the tutorial that walks through it.