MCP flakiness destroying the utility of scheduled tasks (Chrome and Todoist mainly)

Hi all,

Finally getting to the point that my robot assistant is starting to earn its keep. I’ve spent a LONG time setting it all up. I think, for now, the robot assistant is very much in the ‘homekit/home automation’ space. Us nerds are ok with the endless babysitting and fiddling these things require, but it’s a long way from mass market adoption. IMHO, of course.

Anyway, to business. Everything is working pretty well in terms of skills, with the exception of Chrome. I’ve got Claude running my CRM for me. Absolutely incredible (when it works). It updates last contacted, gives a context to all contacts, gives me insights etc etc.

However, I do a lot of my networking on Linkedin (I know, I know, but it’s useful). Linkedin jealously guards their platform, they don’t want you going off-piste (social network, innit?) and using the broader web, they won’t allow Claude to plug in via MCP or anything. So I have Claude trawl my conversations (read only, of course) and update my CRM with anything relevant.

The only problem is, that Chrome and Claude are more fragile than a . The connection breaks every day. Completely destroys this and a few other tasks that rely on Claude using Chrome.

Is there any workaround to this? A better MCP? Getting my data out of Linkedin without doing a mass export?

All advice gratefully received (unless it’s ‘don’t use Linkedin’ - I don’t have much of a choice in this).

Thanks all

I’m really sorry that I cannot be more clear, but I’m working from memory…

A few weeks ago I read something on the internet where the blogger was using a lesser known chromium based browser to circumvent the limitations of websites to write from an ai controlled browser.
Chromium based to make it work via the claude-chrome MCP
And a niche browser because it’s not recognized by the websites. And the blogger had tested various browsers.
Problem: this is all I can remember and I don’t know which blog it was, which browser he used and whether or not it will work for LinkedIn. And I can’t find it anymore.

Hope this helps a tiny bit. :smirk:

1 Like

MCPs are not your friends. They are paywalls, with the payment going to Anthropic for the privilege of spending a significant overhead in tokens to use your own data. If the information you are dealing with is static, then I’d extract it and let Claude deal with it locally. You save money, save time and the resultant processes run more reliably.