Lately I’ve been testing both CLI and MCP in Claude Code... and honestly, the CLI wins I list all my installed tools (vercel, supabase, gh, etc.) in CLAUDE .md and let Claude use them to check logs, run queries, or tweak configs. Setup is simpler, auth is native, and observability is crystal clear. There’s still debate on CLI vs MCP, but in my workflow, CLI is miles ahead

Nov 5, 2025 · 1:36 PM UTC

26
4
6
192
Replying to @dani_avila7
cli always wins. github.com/steipete/mcporter

GitHub - steipete/mcporter: Call MCPs via TypeScript, masquerading as simple TypeScript API. Or...

Call MCPs via TypeScript, masquerading as simple TypeScript API. Or package them as cli. - steipete/mcporter

github.com
2
1
22
Replying to @dani_avila7
I’m glad I’m not the only one who wrote off MCP for most things… tho I did give this mcp exec pattern a try and got good results with it… still it feels like we’re getting yet another abstraction wrapping mcp that wraps http or the likes when we could just use cli … still it was fun to try out. github.com/ipdelete/mcp-code…

GitHub - ipdelete/mcp-code-execution: TypeScript implementation of the code execution pattern for...

TypeScript implementation of the code execution pattern for Model Context Protocol (MCP) - ipdelete/mcp-code-execution

github.com
Replying to @dani_avila7
Add context7 into the mix! I haven't bothered with MCP or API yet but I send it to search context7 for the latest source of truth.
Replying to @dani_avila7
💯
"Debt crisis" (2025, 80 x 100 cm) This painting is about the debt crisis in the US and the coming devaluation of the dollar. People all over the world, both poor and rich, will lose their savings. This is the inevitable consequence of the US leftist economic policy. When the king runs out of money, he takes it from his vassals.
277
1,314
54
10,552
Replying to @dani_avila7
When you say auth is native do you mean CC inherits your context and auth or do you somehow auth it properly? I let CC inherit my auth for AWS/Azure not worked out what it takes for agent to be properly authenticated (pass MFA as per CAP conditional access policies, I pre-set that part)
Replying to @dani_avila7
MCP came about before CLI agents were a thing. I completely agree that using battle-tested CLI tools is both better for agents and for humans monitoring them
Replying to @dani_avila7
wild how we ended up with LLMs watching bash prompts now, next step is Claude handling merge conflicts for fun
Replying to @dani_avila7
CLI always, the creator of Claude Code has said this as well.
3
Replying to @dani_avila7
CLI is much better. I hate MCP.
Replying to @dani_avila7
Clearly, having Claude interact directly through CLI makes automation way more seamless than relying on MCP.
Replying to @dani_avila7
CLI is way better. MCP sucks a big chunk of the context window just start.
Replying to @dani_avila7
That's a smart approach, Daniel! Using CLI sounds like a much better idea, isn't it?
Fighting your Demons is never easy. But it is always worth it.

"Demons" Gear

brothersinarms.shop
19
44
4
433
Replying to @dani_avila7
CLI tools don’t occur your context window.
Replying to @dani_avila7
Yo, sounds like you're deep in dev tool decisions. Btw, tryaskit.com gets you 5 dev interviews to validate ideas – we're at 2/100 before launch if you wanna peek.

Ask It - Stop Building Features Nobody Wants

Talk to 5 developers this week and de-risk your roadmap. We handle the recruiting, screening, and scheduling. You just show up with questions.

tryaskit.com
2
Replying to @dani_avila7
So how does it know WHICH tools to use and HOW to use them?
2
Replying to @dani_avila7
I don't think Claude Code (or similar) is not the main use scenario for MCP. Think of remote, complex applications that were originally designed for human interaction. Otherwise, on your local machine CLI is the way to go.
1
1
Replying to @dani_avila7
Skills v1
1
Replying to @dani_avila7
same here, AGENTS.md for DROID and only browsermcp and Ref MCP

AGENTS.md

AGENTS.md is a simple, open format for guiding coding agents. Think of it as a README for agents.

agents.md
5
Replying to @dani_avila7
How does claude know how to use your installed tools?
2
Replying to @dani_avila7
I just switched my MCPs over to CLIs. Seems MCP is more for external production/security sensitive stuff.
The distribution and spreading of the correct PEACE SYMBOL and its common acceptance among the population can effect very much that is positive and good and lead to a general calming and balance in these difficult and dangerous times.

The Peace Symbol

figu.org
10
1
97
Replying to @dani_avila7
I avoid MCP usually. If not covered by CLI/need a GUI - better to use standalone tools like my Agent Sessions: Codex + Claude + Gemini Unified Session History browser/search/limit tracking/Resume in cwd jazzyalex.github.io/agent-se… native macOS app • open source
1
6
Replying to @dani_avila7
Depend again on MCP, you need to optimize them. Claude Code is deeply optimized for it's own integrated tools. MCP are tools similar. If you want to use them efficiently you will need to prompt Claude. Check Claude system prompt and you will see how they do!