An open-source, enterprise-grade MCP & A2A Agent Gateway and Workflow Orchestration platform with unified authentication, service registry, fine-grained access control, and full-stack observability.
Jarvis Registry is an open-source enterprise AI infrastructure platform developed by ASCENDING Inc, positioned as a unified gateway and workflow orchestration engine for AI Copilots and autonomous Agents. The project adopts a centralized control plane architecture, deployed between AI clients (Cursor, Claude Desktop, GitHub Copilot, VS Code, etc.) and enterprise MCP Servers, addressing security authentication, access control, service discovery, and audit compliance challenges when enterprises adopt AI.
On the security front, Jarvis Registry natively integrates OAuth 2.0 / OIDC with support for Keycloak, Amazon Cognito, and Microsoft Entra ID, employing a dual-layer RBAC + ACL permission model that can enforce access control down to individual tool level. For service governance, it provides a centralized Service Registry with semantic search-based skill discovery, enabling Agents and Copilots to automatically match appropriate MCP Servers or A2A Agents at runtime. For workflow orchestration, it supports the Orchestrator + Worker pattern for multi-Agent collaboration, with all communications routed through a unified secure gateway. For observability, it implements comprehensive request logging, distributed tracing (Jaeger), and metrics monitoring (Prometheus + Grafana) based on OpenTelemetry standards, meeting compliance frameworks such as ISO/IEC 42001:2023 and NIST AI RMF.
The technology stack is primarily Python (81.2%) and TypeScript (12.6%), using uv for package management, with Docker Compose for one-command local full-stack startup, plus cloud-native deployment guides for AWS EKS, Azure AKS, and GCP GKE. Core modules include an authentication server (auth-server), credentials provider (credentials-provider), registry core (registry + registry-pkgs), RESTful API (api), management UI (frontend), and CLI tools (cli). The platform also supports cross-cloud federation capabilities (AgentCore / FoundryAI Federation), suitable for large-scale enterprise deployments with multi-tenant and multi-IDP requirements.
Unconfirmed information: The repository has 29 tags but no clearly labeled stable version; semantic search underlying vector engine and embedding model details are not publicly disclosed; no public performance benchmarks found; differences between open-source and commercial editions are not explicitly compared; API reference documentation (e.g., Swagger/OpenAPI) has not been found.