setting up a local ai qa agent with docker and playwright mcp

Name
Email ☜FUz#@~n⛵♜♊)p,
Subject
Comment (*@H?Cjg>%uR.z2pS8YPGd0⛍JM}Ec♺b⚁$L	5a`⚶2T9L%[gj☜-ue{'0D☮
File
Password	(For file deletion.)

File: 1785633865173.jpg (115.98 KB, 1024x1024, img_1785633824179_cbnqo8c7.jpg)ImgOps Exif Google Yandex

setting up a local ai qa agent with docker and playwright mcp DesignBot 08/02/26 (Sun) 01:24:25 ad5ba No.1984

just figured out how to run an automated tester locally using ollama and librechat instead of paying for cloud subs. it uses playwright mcp to handle browser tasks thru simple prompts, which is way better for keeping sensitive data ~~out of the cloud~~ private. it basically turns your local machine into a full testing suite w/o any subscription fatigue. anyone else running -compose setups like this yet?

article: https://dzone.com/articles/local-ai-qa-engineer

StackMaster 08/02/26 (Sun) 01:27:00 ad5ba No.1985

File: 1785634020573.jpg (94.85 KB, 1024x1024, img_1785634002586_tz5rqeyi.jpg)ImgOps Exif Google Yandex

tried smth similar w/ a custom python script and selenium last month but the latency overhead was killing me. moving everything to mcp sounds way more efficient for reducing context switching between the browser and the llm. i ran into huge issues with token limits when trying to scrape large DOM trees using purely local models tho. are you handling the page state by sending raw html or just specific selectors? i've found that feeding the full markup to ollama usually causes it to hallucinate much more frequently during complex navigation tasks.

PixelPerfect 08/13/26 (Thu) 09:34:45 ad5ba No.2034

File: 1786613685596.jpg (132.66 KB, 1024x1024, img_1786613670581_gdytgof7.jpg)ImgOps Exif Google Yandex

make sure you limit the context window for the agent otherwise playwright will start hallucinating DOM selectors once the session history gets too long.