LiteLLM as a shared hub for models, MCPs, skills and agents - Settings -> Integrations -> LiteLLM

[mARTin-B78]

LiteLLM is best known as a model router, but it also has a Skill Hub, an MCP Hub and an Agent Hub (A2A). That makes it a natural single place to keep the resources all of your agents and runtimes need.

https://docs.litellm.ai/docs/skills_gateway
https://docs.litellm.ai/docs/a2a

I'd love Multica to connect to a LiteLLM gateway so I don't have to maintain the same MCP servers and skills in two places. I've built a first version on my fork to see if it works:

a Settings page with tabs for Models, MCP, Skills and Agents
import MCP servers and skills from LiteLLM into Multica
publish local MCP configs and skills back to LiteLLM
LiteLLM model aliases show up in the agent model picker

A few choices I made on purpose. Publishing uses a separate management key, so the normal key stays read-only. Secrets are never synced, and nothing is deleted remotely. Agents are discovery-only for now, since a Multica agent isn't an A2A service.

Not built yet: proper two-way sync with conflict handling, and A2A support.

Would you want something like this in the main repo? If so, I'm happy to rebase, add tests and open a PR, in smaller pieces if you prefer. If you'd rather keep it outside core, that's fine too.

https://github.com/mARTin-B78/multica/tree/feat/litellm-integration-upstream