> yes, but clis thus need self-service commands to provide guidance, and their responses need to be optimized for consumption by agents.

MCP vs Agent Skills:

MCPs once configured cost you tokens even when they are not used. Unlike MCPs, skills use progressive disclosure. The AI agent does not load up the entire context, if the skill is not being used.

MCPs will die off mostly for this reason alone.

How does this compare with playwright CLI?

https://github.com/microsoft/playwright-cli

I personally found playwright-cli, and agent-browser which wraps playwright, both more token-efficient than using the raw mcp.

Odd that this article from Dec 2025 has been posted to the top of HN though

Its easier to connect to existing sessions in your main browser.

It’s made by Google and comes with Chrome

> That doesn't solve the issue here because the amount of data in the browser state dwarfs the MCP overhead.

The problem with MCP is that you are paying the price in token usage, even if you are not using the MCP server. Why would anybody want that?

And no, the tool search function recently introduced by Anthropic does not completely solve this problem.

What are the numbers? Are there problems other than context usage you refer to?

Can you elaborate more?

> Also MCP is very obviously dead

Some people will push back on this. They are holding out hope that the recent improvements Anthropic has made in this regard have improved the context rot problem with MCP. Anthropic's changes improve things a little. But it is akin to putting lipstick on a pig. It helps, but not much.

The reason MCP is dying/dead is because MCP servers, once configured, bloat up context even when they are not being used. Why would anybody want that?

Use agent skills. And say goodbye to MCP. We need to move on from MCP.

Is your agent harness dropping the entire MCP server tool description output directly into the context window? Is your agent harness always addig MCP servers to the context even when they are not being used?

MCP is a wire format protocol between clients and servers. What ends up inside the context window is the agent builder's decision.

> it is akin to putting lipstick on a pig. It helps, but not much.

The lipstick helps? This had me in stitches. Sorry for the non-additive reply. This is the funniest way I have seen this or any other phrase explained. By far. Honestly has made my day and set me up for the whole week.

I'm a layman here. How is a skill any better? Aren't agent tools loaded on-demand, just as a skill would be? People are mentioning OpenAPI, but wouldn't you need to load the spec for that too?

i am using notion mcp. is there a corresponding skill. also wtf is a plugin.

The bloat problem is already out dated though. People are having the LLM pick the MCP servers it needs for a particular task up front, or picking them out-of-band, so the full list doesn't exist in the context every call.

I am very very glad that most of the world has moved on from this way of doing things. Such a terrible waste of time on a large scale.

> I do not get what's special about banking apps as opposed to online banking.

I use both. In the beginning I used to prefer the web version. I can use my large monitor to see more data and use a full keyboard and mouse. But I have started to use the mobile version more. For Wells Fargo at least, the mobile version is faster to log into because of face ID support. The website requires a lot more clicks and keystrokes. Also, the mobile app makes it easy and possible to deposit checks if and when I get them.

In this case, it is a little more than anticipation. Software dev job losses have already happened and continue to happen.

But have we seen any that are plausibly caused by AI, rather than just using AI as a cover for more boring, typical business reasons?

> But have we seen any that are plausibly caused by AI..

Does it matter if Atlassian is losing seat licenses because devs are being let go?

If that could be proven, then it would be a plausible cause. But without that detail, this is circular reasoning. Those people were let go for the same reason Atlassian let people go -- not AI, just a cover for more mundane business reasons.

What do people use Atlassian for these days?

Back in the day we used Jira a lot. But we also had GitHub accounts. The integration of the two wasn't great so we switched to GitHub issues and kicked Jira to the curb. This saved money and also a lot of hassle in trying to keep the two in sync. I am wondering if many others went a similar route.

We did use Confluence and I do miss it. I would love to find an opensource replacement for it that stores files in markdown format. Like Obsidian, but for teams.

Other countries have been bombed by Iran in this war and none have actively joined the war.