[guiambros]

I don't understand where you're coming from. From the article: "1Password draws a firm line: we will not use MCP to expose raw credentials or secrets."

That seems a pretty reasonable argument to me. MCP is a complete hack, and the risk of agents going rogue (or getting hacked, or finding some vulnerability, 0-day, etc) and exposing your entire secrets database is just too high for 1P to accept. As a customer for 15+ years, that's exactly what I'd like to hear from my password manager.

Are users soon going to demand a way to give agents access to their passwords? Yes, absolutely.

With "AI browsers" and a whole industry of startups building agents, you can count the months until users start asking their password managers for ways to grant permission.

What 1Password is saying is "fine, but we need to do better", and MCP is an insecure clusterf*.

I think that's very reasonable, although I'll reserve judgement for when they release the so-called "secure agentic access".

[everyyear]

Every single time in the past 10 years that I have seen some software tech hyped here it has been the same. Look under the hood, it's just some over-engineered stab at vendor lock-in.

The tech industry has become a pyramid scheme to sell more computers. Everything feeds into that. "Cybersecurity", "cyberwarfare" scares, which are enabled by the over-engineering, will be solved by? Yeah more over-engineering. A boom of GPU farms where the answer to all shortcomings is using more GPU farms? First "agent"/mcp codebases I looked at were some naive reinvention of expert systems to scaffold and parse prompts and responses from some LLM endpoint.

It's transparent at this point.