February 2026 monthly summary for stacklok/toolhive focusing on delivering robust routing, resource management, telemetry, and reliability improvements. Highlights include enhanced composite tool routing semantics, safer host header handling across deployments, sensible default resource limits for MCP containers, and enriched telemetry middleware with precise request metrics. Expanded test coverage with unit and end-to-end tests, and updated documentation to clarify visibility and routing semantics.

January 2026 focused on stabilizing VMCP/ToolHive configuration, strengthening runtime safety, and boosting observability, with a strong emphasis on maintainability and business value. Key features include automated type coercion after composite tool template expansion to ensure backend validation passes, and standardizing VMCP config field tags to camelCase for Kubernetes-friendly YAML. Foundational work to simplify config management was advanced by moving remaining VMCP CRD fields into spec.config and removing rawConfig in favor of strict unmarshalling. Reliability improvements were delivered via a new recovery middleware that protects the server from panics and preserves service availability. Template validation improvements were implemented to support runtime template functions (e.g., fromJson, json, quote) within composite tools, eliminating validation-friction for common use cases. In parallel, telemetry/config migration fixes and API stability improvements were completed, contributing to safer upgrades and predictable behavior. Business value delivered: more reliable tool execution, safer upgrades, clearer configuration, and better observability with lower maintainance overhead for operators and developers.

December 2025 focused on increasing observability, reliability, and template-driven flexibility for VirtualMCP, delivering practical business value through improved visibility, safer operations, and more resilient workflows. Key deliverables include instrumentation of the Virtual MCP Server with metrics and distributed tracing, enabling telemetry configuration via CRDs, and adding debug logs to differentiate MCP server types. We also implemented graceful shutdown for foreground servers to prevent flaky shutdown behavior. Composite tools gained support for non-string arguments and structured content, with thorough tests and documentation. Workflow defaults and on-error handling were enhanced to improve reliability for skippable steps. The testing framework for VirtualMCPServer was strengthened with mock HTTP, retries, and readiness checks to reduce flaky tests. Minor bug fixes included propagating pod template annotations, removing a large binary that caused CI failures, and adding fromJson support to templates for dynamic data access.