Now I know why no one uses the google cloud. Making TTS and STT working costed me the whole night. Gemini was easy tho. But fuck google, you costed me a lot of energy. You guys are crazy. Now my api connects in a magic way i don't even understand with the gcloud cli app. The rest of my application is totally rest, don't use much of the google library.

I implemented google TTS and STT into ChatGPT. I use for somethings google because it's cheaper. It works using a JBL Go! speaker. I can just turn it on and start chatting with it. I implemented google search and gave it a memory. It can remember numbers for me. It accepts dutch and english. I can say 'google' and google is the main action. It will fetch results from google and uses gpt to summary the results. It works perfect! BUT FUCKING AI. I want to know the color of the hair from Mona Lisa. Not freaking Ona Isa! I send it literally correct. The speechtotext works great. But fucking API with it's reading. Pathethic. How far is AI? Barely usable as home assistant. So far - besides auto completion and giving code snippets / concepts AI is freaking useless. You need more patience for AI than a kid.

I hope the inventor of oauth2 dies alone. He should.

Lately, I’ve been diving deep into Gemini CLI and experimenting with multiple MCP integrations. I started by connecting Context7, which fetches real-time, version-specific docs and code snippets right into your prompt — super useful for quick reference and building faster.

Then I added Taskmaster-AI, which lets me tap into both OpenAI and Google API keys. This unlocked a much broader range of language models for different use cases like planning, summarizing, or automation.

To top it off, I included SmolAgents, a lightweight agent framework from GitHub that allows for modular, goal-driven AI behavior directly in the terminal.

What surprised me is how seamlessly these all work together inside Gemini CLI. You don’t need to call them out directly — once configured properly in your .gemini/settings.json, they complement each other naturally. Taskmaster handles structured planning, Context7 brings up-to-date references, and SmolAgents give flexible control.

I’m curious if anyone else is combining tools like these in their workflow? What kinds of prompts or workflows are you using Gemini CLI for?

Also, if you're interested, I made a quick walkthrough, check out the link in my profile.