Editing Blog/2026-03-02/Copy To Claude

One of the things I find annoying these days is not being able to unleash Claude Code on a problem and let it spin. You have to copy out a cookie and instruct Claude on how to get going on the website and do work and so on. So I thought of a fairly simple idea that has got me a lot of mileage. I added a Copy To Claude button to the micro-SaaS I run for my clients that creates a short-lived token and a simple prompt that describes the page. Surprisingly, something this simple actually allows me to navigate to a random page in the UI, then hit Copy To Claude, and then paste it into Claude Code to start using programmatic navigation. 

Since I usually do roles and permissions, I can make the token generated there be a short-term access token with lower privileges than me so the program can effectively navigate everything but not do anything damaging. Or I could choose to navigate to an appropriate page and get a Copy To Claude that is more powerful.

To increase the functionality of an LLM on the app, these days I write most in a server-side rendered language so that the user-agent (either the browser or the LLM) only sees HTML. The thing I like most these days is HTMX because it's fast, easy for LLMs to write and style, and because you get back raw HTML even on sub-component AJAX views. This allows a fully LLM navigation mode. I think this style of website will probably start becoming more popular as people try to integrate with their users' agents.

One day, I hope to either WebMCP this website or to serve a [[wikipedia:Gemini (protocol)|Gemini Protocol]] version of it through both <code>gemini://</code> or a HTTP proxy. With an appropriate embedding index and vector search, it should be easy-peasy for people to search my site with their agents without ever having to visit it!

<syntaxhighlight>
You are viewing the Technology Brother Data Dashboard (powered by Skybright).

## Current Page
Pipeline detail page for "skybright_ga4adsclientdata__xxx_great_mood" (ID: 12345678-32f1-4635-9a60-abcdef12345). Destination dataset: skybright_ga4adsclientdata__xxx_great_mood, working dataset: skybright_ga4adsclientdata__xxx_great_mood_googleanalyticsdataapi_0849. Shows basic info, datasets, table freshness, associated pipeline request, 1 connection(s), recent syncs, encrypted credentials, and manual actions (report failed sync to Fermat).

## How to fetch this page
curl -s -b 'access_token=Bearer eyXxx1234' 'https://admin.demo.example.com/pipelines/12345678-32f1-4635-9a60-abcdef12345'

Note: The response is HTML. You can pipe through html2text or parse with BeautifulSoup/pup.

## Available Navigation
- Dashboard: curl -s -b 'access_token=Bearer eyXxx1234' 'https://admin.demo.example.com/dashboard'
...

</syntaxhighlight>

[[Category:Blog]]
[[Category:AI]]
[[Category:Tips]]