MCPHub
Back to changelog
v1.2.0Nov 1, 2026

v1.2.0

v1.2.0 introduces experimental cluster support through an opt-in PostgreSQL multi-replica mode. Multiple MCPHub instances can share configuration and credentials behind a load balancer, with configuration convergence, cross-node OAuth coordination, readiness checks and graceful HTTP draining. Docker Compose and Kubernetes examples help you evaluate the new deployment model. This release also improves personal credential handling and restores generic OIDC login with Better Auth 1.7.

View on GitHub

Summary

v1.2.0 introduces experimental cluster support through an opt-in PostgreSQL multi-replica mode. Multiple MCPHub instances can share configuration and credentials behind a load balancer, with configuration convergence, cross-node OAuth coordination, readiness checks and graceful HTTP draining. Docker Compose and Kubernetes examples help you evaluate the new deployment model. This release also improves personal credential handling and restores generic OIDC login with Better Auth 1.7.

Highlights

  • Experimental cluster / multi-replica mode (disabled by default): enable MCPHUB_REPLICA_MODE=true to run multiple instances against shared PostgreSQL. Configuration changes converge every five seconds, stale configuration writes are rejected, and OAuth state, authorization-code consumption and credential refresh are coordinated across nodes.
  • Deployment and operations: Docker Compose and Kubernetes examples include separate schema setup, two replicas and load-balancer affinity. Liveness/readiness checks, node/configuration diagnostics, metrics and graceful HTTP draining help operate replicas. Use shared persistent JWT and credential-encryption keys, run schema setup separately, and set DB_SYNCHRONIZE=false on serving nodes. See the replica deployment guide.
  • Experimental boundaries: MCP sessions, isolated upstream clients, cookie jars and local files remain process-local. Stateful clients require sticky routing and reinitialization after node loss; in-flight calls are neither migrated nor replayed. PostgreSQL and load-balancer high availability remain deployment responsibilities. Evaluate in a test environment before adopting this mode in production.
  • Add documentation for the community-maintained Home Assistant app.

Fixes

  • Restore generic OIDC login with Better Auth 1.7.
  • Persist upstream OAuth expiry and renew personal tokens before use.
  • Show safe personal credential previews for administrators.

References

v1.2.0 Changelog · MCPHub