MCP Hangar -- Control

MCP Hangar -- Control Plane for MCP Servers is an open-source, self-hosted control plane for MCP servers. It sits between AI agents and MCP servers and is described as a place to decide who can call what, see what is happening, and prove it to auditors. The page also says no cloud account is required to get started.

Its stated capabilities center on governance, observability, reliability, and compliance. Governance features include per-caller access policies, per-server and per-tool RBAC, tool access filtering by glob pattern, authentication with API keys, JWT/OIDC, and identity propagation to audit logs. A mutator pipeline supports priority-ordered payload transformations and an audit-only shadow mode. For observability, it provides full call logging with caller identity, server, timestamp, and arguments, along with OpenTelemetry and Prometheus support, cost attribution, and an interceptor framework based on hooks and wildcard subscriptions. Reliability features include parallel execution across servers, backpressure and fair scheduling, a circuit breaker, server groups with load balancing, failover, health tracking, and lifecycle management with lazy loading and automatic shutdown after an idle TTL. Compliance-oriented functions include an identity-aware audit trail, SOC2 evidence generation, EU AI Act readiness, SIEM export, digest pinning for tool schemas, and multi-server orchestration across local, Docker, and remote HTTP backends.

The quick-start section says the agent installs from PyPI with pip or uv and works on macOS, Linux, and Windows. It also says the wizard detects runtimes, can pick MCP servers, and configures Claude Desktop automatically. The page names Claude, GPT, and custom agents in its overview of how it works, and it says the platform offers a unified interface for heterogeneous backends.

MCP Hangar v1.4.0 is released under the MIT License. The page describes the agent as open source and notes that it can be run locally, on servers, or in Kubernetes.