LLMs happened, and then major providers offered API access and many SaaS were born as wrapper companies.

Architecture: User <-> SaaS <-> LLM

This is well and good.

Now, many LLM providers are providing more and more full-featured (and agentic) front-ends for their users. Include claude-code and gemini-cli in this list. These front-ends are also, in general, mcp clients.

Yesterday, Nia posted ( https://news.ycombinator.com/item?id=44671601 ) a developer-focused context-as-a-service subscription MCP server at $15 / Mo.

Architecture: User <-> LLM <-> CaaS

My questions:

1. Are there many known instances of CaaS providers out there? EG, revenue generating products whose interactions w/ end users are primarily (or maybe entirely) mediated by LLM providers?

2. How far can the model go in sufficiently general cases? In the specific Nia case, if its secret-sauce is effective for general software development, I would expect that anthropic and google would seek to embed the same functionality into their agentic coding assistant environments. More niche products, or products that pull from private data sources, seem to me more likely to have at least medium-term viability

3. Would you pay for MCP to augment your agents in personalized ways? A pitch: progressive language-immersion via LLM clients. Combine a traditional interactive-tutoring-system or SRS with MCP hooks such that the agent keeps an up to date handle on the foreign-language concepts you understand well, and your zone-of-proximal-development. It can then communicate by default in your target language at appropriately challenging level and fallback to English where you aren't well equipped.

(the pitch is related to my work with http://github.com/patched-network/vue-skuilder , a general toolkit for SRS++ edtech development)