3 things service catalog does:

* index *

It indexes the whole codebase, and per repo to make a sense of it. 2 types of indexing we are looking at, a codebase level where we let the agent figure out the relation between repositories and per repo, where we extract the data (libraries, dependencies, etc) with respect to what we know of the codebase

* search *

After indexing the whole codebase, we are able to run search. the search can be either lexical or logical. we pass the context of the codebase we generated in the index phase to the search and let search decide what to look for. User will be able to search by asking "find me repositories that are written with python, use library X and all the repos that are relevant to these repos".

* batch change *

Batch change lets users make changes in many services at the same time. So a user will be able to say "Find all repos with github actions job name X and update the version from 1 to 2". The mcp server will fire up claude code in clones targeted repositories, and make a change while passing the codebase and repostiories context to it for better decision making.

After making these changes, user can ask the server to make PRs, and check the PR status of them and give a summary

=================

You can find a video of me using the mcp server here: https://infraas.ai