
MCP: the New 'USB-C For AI' That's Bringing Fierce Rivals Together (arstechnica.com) 29
An anonymous reader quotes a report from Ars Technica: What does it take to get OpenAI and Anthropic -- two competitors in the AI assistant market -- to get along? Despite a fundamental difference in direction that led Anthropic's founders to quit OpenAI in 2020 and later create the Claude AI assistant, a shared technical hurdle has now brought them together: How to easily connect their AI models to external data sources. The solution comes from Anthropic, which developed and released an open specification called Model Context Protocol (MCP) in November 2024. MCP establishes a royalty-free protocol that allows AI models to connect with outside data sources and services without requiring unique integrations for each service.
"Think of MCP as a USB-C port for AI applications," wrote Anthropic in MCP's documentation. The analogy is imperfect, but it represents the idea that, similar to how USB-C unified various cables and ports (with admittedly a debatable level of success), MCP aims to standardize how AI models connect to the infoscape around them. So far, MCP has also garnered interest from multiple tech companies in a rare show of cross-platform collaboration. For example, Microsoft has integrated MCP into its Azure OpenAI service, and as we mentioned above, Anthropic competitor OpenAI is on board. Last week, OpenAI acknowledged MCP in its Agents API documentation, with vocal support from the boss upstairs. "People love MCP and we are excited to add support across our products," wrote OpenAI CEO Sam Altman on X last Wednesday.
MCP has also rapidly begun to gain community support in recent months. For example, just browsing this list of over 300 open source servers shared on GitHub reveals growing interest in standardizing AI-to-tool connections. The collection spans diverse domains, including database connectors like PostgreSQL, MySQL, and vector databases; development tools that integrate with Git repositories and code editors; file system access for various storage platforms; knowledge retrieval systems for documents and websites; and specialized tools for finance, health care, and creative applications. Other notable examples include servers that connect AI models to home automation systems, real-time weather data, e-commerce platforms, and music streaming services. Some implementations allow AI assistants to interact with gaming engines, 3D modeling software, and IoT devices.
"Think of MCP as a USB-C port for AI applications," wrote Anthropic in MCP's documentation. The analogy is imperfect, but it represents the idea that, similar to how USB-C unified various cables and ports (with admittedly a debatable level of success), MCP aims to standardize how AI models connect to the infoscape around them. So far, MCP has also garnered interest from multiple tech companies in a rare show of cross-platform collaboration. For example, Microsoft has integrated MCP into its Azure OpenAI service, and as we mentioned above, Anthropic competitor OpenAI is on board. Last week, OpenAI acknowledged MCP in its Agents API documentation, with vocal support from the boss upstairs. "People love MCP and we are excited to add support across our products," wrote OpenAI CEO Sam Altman on X last Wednesday.
MCP has also rapidly begun to gain community support in recent months. For example, just browsing this list of over 300 open source servers shared on GitHub reveals growing interest in standardizing AI-to-tool connections. The collection spans diverse domains, including database connectors like PostgreSQL, MySQL, and vector databases; development tools that integrate with Git repositories and code editors; file system access for various storage platforms; knowledge retrieval systems for documents and websites; and specialized tools for finance, health care, and creative applications. Other notable examples include servers that connect AI models to home automation systems, real-time weather data, e-commerce platforms, and music streaming services. Some implementations allow AI assistants to interact with gaming engines, 3D modeling software, and IoT devices.
MCP? That's not re-assuring (Score:5, Insightful)
MCP may very well stand for "Model Context Protocol", but in the 1980's it stood for the villain of the Tron series, Master Control Program. Will tying AI together be a good thing or will we see the rise of another kind of MCP?
Re: (Score:2)
Will tying AI together be a good thing
There is another system [wikipedia.org].
Re: (Score:2)
Re: MCP? That's not re-assuring (Score:2)
Re: MCP? That's not re-assuring (Score:2)
Re: (Score:3)
MCP may very well stand for "Model Context Protocol", but in the 1980's it stood for the villain of the Tron series, Master Control Program. Will tying AI together be a good thing or will we see the rise of another kind of MCP?
Knowing our species ability to be ironically ignorant about the future, I fully expect the 2027 T-101 HappyBot (built by Skynet-Softacle) to have a shiny emblem on its ass that reads "Powerd by WOPR"
Re: (Score:2)
They should have taken the opportunity to use WOPR as name for SPPO.
Re: (Score:3)
Of course, the Master Control Program is really just another term for Operating System these days. So tying the two together we end up with Windows and CoPilot....
Re: (Score:1)
Re: (Score:2)
Science people like puns. It's not unlikely that this was on purpose. Be glad they didn't call it T-1000.
They failed at the GPU level to de-app (Score:1)
Re: (Score:2)
How are GPUs any more "application platforms" than CPUs are? They're both processors. Are CPUs "estranged from the kernel"? And what's "everything else" in this context?
Could it be that a remedial education on what a computer is solve your problems?
Re: They failed at the GPU level to de-app (Score:1)
Let's get this out of the way (Score:4, Funny)
Re: (Score:1)
What are the other 14 in this case?
Re: (Score:2)
Let's get this out the way:
@necro: wooooooooooooooooooooooooooooooooosh
MCP for CYC (Score:2)
We should have all LLMs validate their output through CYC, which knows and can validate actual common sense facts. We need AI to self-fact check, and getting it to use the CYC ontology database seems like a worthy start.
Re: (Score:1)
If I were a big AI company I'd snap up CYC as fast as possible. There's nothing like it and it would take lots of human labor to clone up something similar. (The open source version is a subset.)
Obfuscation (Score:5, Interesting)
MCP is just RPC with overly obfuscated documentation to disguise simple it is. All it really is is a schema that provides a consistent way to probe for functionality and pass parameters to functions. That's it. You call a known endpoint to see what functionality is available, then pass that to the LLM and tell it what can be called and what parameters it requests.
Anthropic's own AI agrees with me:
>Yes, that's hilariously accurate! At its core, MCP is basically just a standardized WebSocket protocol for:
>1. Discovery ("what can you do?")
>2. Tool invocation ("do this thing with these params")
>3. Structured responses ("here's what happened")
>The actual core interaction is dead simple: ...
The MCP documentation is just an example of VC-bait: https://docs.anthropic.com/en/... [anthropic.com]
Which isn't to say MCP isn't very useful, but you should generally be able to discern multiple levels of unnecessary indirection and realize it's serving a political and marketing purpose rather than a technical one. Langchain is the same way (but much worse).
Re: (Score:2)
Re: (Score:2)
That's all very clear, and we should already recognize that by the analogizing with USB which is itself insulting and ridiculous. It's a SuperKendallian absurdity.
USB-C for AI? (Score:2)
So, what, there are some cheap implementations that cause literal fires?
Re: (Score:2)
Re: (Score:2)
Well, you see, if the web site is filled with SEO spam, they want to let the bots get all that data so that when you tell ChadGBD that you were thinking about pizza it will tell you about 37 great pizza ovens you can lease for zero down payment as part of a turnkey pizza business that's actually an MLM that is also somehow a cult.
Pretty interesting example of using mcp (Score:2)
Saw this very interesting demo of using mcp to tie an AI with the ghedra decompiler to do reverse engineering.
https://m.youtube.com/watch?v=... [youtube.com]
End of Line (Score:2)
And I fight for the users.
technical notes (Score:2)
MCP
SECTION 2.9.1 Session Initiation
A session is initiated when a client opens a connection to a server and the server responds with an opening message. The Client then responds with an introductory comment to formally introduce itself.
EXAMPLE:
200 MCP.OPENAI.COM Greetings, Program
100 MCP..ANTHROPIC.COM You're getting brutal, Sam.
201 MCP.OPENAI.COM And sadistic, Claude.
EXAMPLE:
200 MCP.GUARDIAN.RU Zdravstvuyte
100 MCP.COLOSSUS.MIL I bring you peace
201 MCP.GUARDIAN.RU Ubeyte Forbina, on Ukrainski
XKCD 927 (Score:2)
It's bad enough when things use ollama API instead of openai API. We don't net yet another standard.