samanhappy/mcphub v1.0.35

on GitHub

Summary

This release hardens MCPHub's reliability and interoperability. Upstream OAuth servers now reconnect cleanly after reauthorization, and on-demand server settings (startOnDemand/idleTimeoutMs) are correctly persisted when running in database mode. Multipart file uploads are decoded and validated more robustly, and embedding dimensions are only forwarded to models that support them. Smart-routing configuration has also been modernized to provider-neutral field names with automatic migration of legacy persisted values, while existing OPENAI_* environment aliases keep working.

Features

• use provider-neutral smart routing config names by @atirna in #1133

Fixes

• persist startOnDemand/idleTimeoutMs when running in database mode by @Rahulsharma0810 in #1095
• reconnect upstream OAuth servers after reauthorization by @samanhappy in #1128
• decode multipart binary file arrays by @emecii in #1130
• only forward embedding dimensions to MRL-capable models by @samanhappy in #1132
• reject multipart content type line breaks by @emecii in #1137

摘要

本版本提升了 MCPHub 的可靠性与互操作性。上游 OAuth 服务器在重新授权后现在可以干净地重连；按需启动的服务器设置（startOnDemand/idleTimeoutMs）在数据库模式下也能被正确持久化。多部分文件上传的解析与校验更加健壮，嵌入向量维度只会转发给支持它们的模型。智能路由配置也已统一为供应商中立的字段命名，并自动迁移历史持久化值，同时保留现有的 OPENAI_* 环境变量别名。

功能

• 使用供应商中立的智能路由配置命名 by @atirna in #1133

修复

• 在数据库模式下运行时持久化 startOnDemand/idleTimeoutMs by @Rahulsharma0810 in #1095
• 重新授权后重连上游 OAuth 服务器 by @samanhappy in #1128
• 解码多部分二进制文件数组 by @emecii in #1130
• 仅向支持 MRL 的模型转发嵌入向量维度 by @samanhappy in #1132
• 拒绝多部分内容类型中的换行符 by @emecii in #1137

New Contributors

• @emecii made their first contribution in #1130
• @atirna made their first contribution in #1133

References

• use provider-neutral smart routing config names by @atirna in #1133
• persist startOnDemand/idleTimeoutMs when running in database mode by @Rahulsharma0810 in #1095
• reconnect upstream OAuth servers after reauthorization by @samanhappy in #1128
• decode multipart binary file arrays by @emecii in #1130
• only forward embedding dimensions to MRL-capable models by @samanhappy in #1132
• reject multipart content type line breaks by @emecii in #1137
• refactor: remove dead code and logging from mcpService.ts to resolve CodeQL alerts by @samanhappy in #1124
• docs: prefer gh api for security advisories by @samanhappy in #1125
• docs: prefer gh CLI for GitHub operations in agent guides by @samanhappy in #1136
• Full changelog: v1.0.34...v1.0.35