eoskx

Claude's style is now overuse of sentence fragments, which this "article" has in spades. Fragments are the new emdash or "delve".

AI generated slop. Why would anyone waste the time to read this if you won't take the time to thoughtfully write something?

It is unfortunate because the content and subject matter isn't actually bad, but yes, there are definite AI-generated tells here.

A lot of how well it works or won't work depends on your clients, as not all clients have support for things like RFC 9728 (Protected Resource Metadata). Assuming your client has good support for most of the OAuth 2.0 standards that MCP uses (you don't need DCR as you can statically register clients, assuming that is viable for your environment), then it is possible now with most IDPs to get an OAuth 2.0 auth code flow working just fine. You would then do a token exchange to the upstream to ensure to get the appropriate new audience and rescope/downscope as necessary. Gateways can also help here a lot as instead of baking in all of the auth concerns into your MCP Servers and upstreams, you delegate that to the MCP Gateway. Again, gateway here means different things to different vendors, but Kong, for example, has the ability to act as an MCP proxy (gateway), expose tools based on consumer role or group, apply OAuth 2.0 to it and do an upstream token exchange, while also acting as a regular API gateway that can protect an endpoint with OAuth 2.0/OIDC.

Doing a workshop this week on MCP for an enterprise client and explaining the 406 returned by GET against /mcp w/o text/event-stream is exactly one of the things that I have to bring up when I do these.

The specification still leaves a lot to be desired, especially as it relates to auth. There are lots of bad ways to do auth with MCP and only a couple of good ways. It also puts a lot of pressure on the various IdP vendors and relies on lesser used areas of OAuth 2.0/2.1 (like DCR, token exchange, etc.). It started out in a place where the assumption was you were running an MCP Server on a laptop or you were a SaaS provider serving lots of individual users -- somehow DCR in the initial spec iteration seemed like a good idea (spoiler: it wasn't) and fortunately, the latest revision has somewhat addressed that. XAA/ID-JAG & CIMD should continue to round-out some client management and auth solutions for the enterprise.

Gateways are another area that needs to be addressed in the spec. There isn't a formal definition of one in the spec, and yet, there are lots of "gateways" out there. What a gateway is and what it should do is an open question and it means different things to different people depending on who you ask. For example, who does token exchange: the MCP server or the MCP gateway? Both are valid answers right now depending on the implementation or your opinion of which is best.

More spec iterations should be coming this year. I'm still pretty optimistic about MCP as a whole, as it remains a good way to standardize tool calls across agents and some of the other entities that it provides like resources and prompts are genuinely useful to add more determinism to an agent. Interceptors and skills will be good, too.

If you're interested in helping to evolve the spec further, the MCP Contributors Discord is active. There are lots of IGs/WGs that solicit feedback and you can participate in meetings with your feedback.

Fond memories of buying cheap Sun gear around 2005-2007. I had an E4500, Blade 1000, and a Tadpole SPARCbook 6500 that I ran Solaris 10/11 on along with a couple of Sun Rays. Used the Blade 1000 as a Sun Ray server and it was a great experience. Glad to see it is still alive and kicking in some form.

Interesting, but cannot run CUDA or more to the point `nvidia-smi`.

Also, not surprising that LiteLLM's SOC2 auditor was Delve. The story writes itself.

Valid, but for all the crap that LangChain gets it at least has its own layer for upstream LLM provider calls, which means it isn't affected by this supply chain compromise (unless you're using the optional langchain-litellm package). DSPy uses LiteLLM as its primary way to call OpenAI, etc. and CrewAI imports it, too, but I believe it prefers the vendor libraries directly before it falls back to LiteLLM.

Not just as a gateway in a lot cases, but CrewAI and DSPy use it directly. DSPy uses it as its only way to call upstream LLM providers and CrewAI falls back to it if the OpenAI, Anthropic, etc. SDKs aren't available.

Yep, I think the worst impact is going to be from libraries that were using LiteLLM as just an upstream LLM provider library vs for a model gateway. Hopefully, CrewAI and DSPy can get on top of it soon.

Yep, DSPy and CrewAI have direct dependencies on it. DSPy uses it as its primary library for calling upstream LLM providers and CrewAI falls back to it I believe if the OpenAI, Anthropic, etc. SDKs aren't available.

LangChain at least has its own layer for upstream LLM provider calls, which means it isn't affected by this supply chain compromise. DSPy uses LiteLLM as its primary way to call OpenAI, etc. and CrewAI imports it, too, but I believe it prefers the vendor libraries directly before it falls back to LiteLLM.

This is bad, especially from a downstream dependency perspective. DSPy and CrewAI also import LiteLLM, so you could not be using LiteLLM as a gateway, but still importing it via those libraries for agents, etc.