In its simplest form, it's just a dump of the code intelligence information from a static copy of the code. This can power an LSP, however, without additional logic wouldn't be able to handle a project under edit, since the locations won't match between the indexed state and the edited project state; So it lends itself well for something like Sourcegraph that already displays a static copy of the codebase.

Uber uses SCIP as part of the LSP implementation for our Java monorepo (Pieces of which we've [open-sourced](https://github.com/uber/scip-lsp)). Standardizing on SCIP has helped us generalize tools to be independent of the compiler/language ecosystem (eg we could do call-stack-analysis on any project that exports valid SCIP; do feature flag cleanup; find refs/impls across a wider scope than most LSP servers can handle due to memory constraints).