[CONTEXT]

we are a digital agency that relies on Hive daily. We've been trying to make the MCP IA integration work reliably with Claude since early May. What follows is a brief and only mildly bitter summary.

[SECURITY PROBLEM]

The connector requires embedding an API key directly in the URL. Since Claude.ai shares connectors org-wide, this meant our director-level key was accessible to every team member — a significant privacy and security issue. Your server supports OAuth 2.0, which would solve this entirely, but Claude.ai's connector UI doesn't support OAuth discovery yet. So our workaround? We hired an additional Hive seat just to create a restricted account with a safer key. In 2026, we paid for an extra license to work around an authentication model that belongs in 2018.

[STABILITY PROBLEM WE CAN´T WORK AROUND]

Even with that workaround in place, the connector has now stopped loading its tools altogether in Claude.ai sessions. We're seeing this regularly:

* getProjects failing while other calls succeed

* Full MCP session dropping mid-workflow

* Inconsistent tool availability across sessions

* Generic error messages that provide zero debugging value

* insertActions working while getProjects fails — which is a special kind of chaos

[MID-2026 AND STILL NOT PRODUCTION-READY]

MCP as a technology has been heavily pushed by Anthropic and the broader AI ecosystem since late 2025. It's been positioned as the standard for AI-tool integration going forward. We adopted it early, invested time, documented workarounds, opened support tickets, bought extra seats — and today, in the middle of 2026, our most critical business integration still can't reliably load its tools in a web session.

There is a real possibility that by the time this is fully stable, the ecosystem will have moved on to something else entirely. That would be a remarkable achievement.

What we're actually asking for

* Investigate why the connector is failing to load tools in Claude.ai right now — this is a blocker for our daily operations.

* Prioritize OAuth support for claude.ai so we can stop paying for a workaround seat.

* Improve error messages and session diagnostics so failures are actually debuggable.

* Consider project-scoped API keys as a mid-term solution.

We genuinely like Hive and believe in the potential of this integration. We just need it to work.