@chip.gpt: How to identify good vs bad MCP servers. Consumers should be critical of the tools they give their AI. Developers should be mindful of the tools they produce. #ai #openai #mcp #llm #softwareengineer
It is not clear to me what problem you’re trying to solve. MCP defines a protocol that is supposed to dynamically allow discovery of tools. You are basically re-implementing that adding some guided behavior. However, doing so you are also preempting the model from making some out decisions so I’m not sure how valuable this approach is. Do you have data to show it is effective
2025-08-19 20:53:17
7
Vinnie :
How is the client going to become aware of the updated tools list? Many clients just cache the tools list at the start
2025-08-15 02:16:11
3
ELIS KUCEVIC :
doesn't this add significant latency?
2025-08-14 16:16:12
3
macoveiaf :
after watching yoyr demo video and using my very limited expertise, this sounds like a good idea for developer facing agents, but not that great for customer interfaces. i dont want my clients to have to navigate through tools and i wouldnt trust the llm to correctly navigate the nested structure and call the correct tools on its own either. any thoughts on this?
2025-08-15 16:03:21
2
Landon🏡🌱🌻💾🎸 :
what programming tools do you use? for example, do you recommend a python or typescript library?
2025-08-15 00:25:49
0
Oliver Taylor :
Ok this makes MCP seem like more than a redundant API layer.
2025-08-15 00:52:51
5
Abir :
I saw this in the MCP Reddit today, was that you??
2025-08-15 00:12:00
4
Lan 🌊 AI startup friend :
Thank you! Redesigning my mcp server now
2025-08-14 22:45:27
7
NJK | Lock In 🐺 :
Image you have Gmail MCP, Salesforce MCP , Jira MCP ! Just image de LATENCY of this architecture 👀👀😭😭😭😭
2025-08-21 01:43:00
1
MakiabelMFE :
1:35 is when he stops complaining and starts getting to the point
2025-09-14 11:11:26
2
Smol :
I want to make an MCP for a “Factory as a Service” which gives lllms the ability to make, order and ship from a suite of quality print-on-demand products.
2025-08-15 00:16:40
0
zzzzzzzzzzzzz :
I am working on solution for this
2025-08-15 05:39:41
1
sonicpoet :
Just like “press 3 for accounting”
2025-10-12 12:29:46
0
keldon :
You’ve given me something to think about. I need to corroborate a few things you’ve put forth here.
2025-10-04 18:52:26
0
TheSethRose :
Bro - GitHub MCP 😬
2025-08-14 23:56:09
1
websites.carol :
Amazing thanks
2025-10-04 21:14:58
0
paul_dentro :
this is also very useful concept when thinking about adding tools to an LLM. Never thought about nesting tools and exposing only their parents meaning at first, smart!
2025-09-26 03:30:23
0
RocketJoe :
is this tuff in america?
2025-10-02 12:46:51
0
mediadada :
Doesn’t this add round trips between the client and the MCP server? Maybe that’s fine, but just wondering.
2025-10-06 03:52:12
0
TechZik :
Cool let me vibe that to my website test tool
2025-12-29 05:36:34
0
icoveredthatsong :
How does the client experience differ?
2025-10-11 13:19:13
0
Prince :
How does the approach for that multiple nested llm call with toolcall? Isnt that weird
2025-10-25 01:23:10
0
Derek :
so server returns different tools when listed based on a session id?
2025-08-15 05:46:57
0
To see more videos from user @chip.gpt, please go to the Tikwm
homepage.