How do you use this across multiple projects

[menacher]

UI, API, DB and other projects are usually in different folders, is there a way to init this graph across multiple projects/folders?

[siathalysedI]

I have a similar issue since I work with a "root project" which has legacy monolith, new-frontend, new-backend, legacy-sql, new-sql, reference-projects. How could I manage this? Since I have also md guides which have references to code? Whenever I do index, it creates some kind of loop I think

[chevyphillip]

These are all valid questions and real-world architectures. What I'm doing for this is using Augment Code VSCode extension or the Auggie Context MCP server, which is built for these exact purposes. I would combine Augment Code with CodeGraph to really unlock these multi-context layers.

[rickross]

I have added support for multiple projects in my fork of codegraph.

[colbymchenry]

Hey everyone — this is now fully supported! Every CodeGraph MCP tool accepts an optional projectPath parameter for cross-project queries.

How it works

1. Initialize each project separately:

   codegraph init -i /path/to/frontend
   codegraph init -i /path/to/backend
   codegraph init -i /path/to/shared-lib

2. Query across projects — Claude can use the projectPath parameter on any tool to reach into another project's graph:

   codegraph_context(task: "auth flow", projectPath: "/path/to/backend")
   codegraph_callers(symbol: "UserService", projectPath: "/path/to/frontend")
   

3. Auto-sync works per-project — the Claude Code hooks (mark-dirty / sync-if-dirty) walk up from the edited file to find the nearest .codegraph/, so each project in a monorepo stays independently synced.

Each project gets its own .codegraph/ directory with its own index. Claude automatically knows to use projectPath when you reference another project that has CodeGraph initialized — no extra configuration needed.

@siathalysedI For your monorepo with multiple sub-projects: initialize CodeGraph in each sub-project root (not the monorepo root), and make sure your legacy/reference directories that shouldn't be indexed are excluded via the .codegraph/config.json ignore patterns. This avoids the loop issue you mentioned.

@rickross We've since added full multi-project support upstream — all MCP tools support cross-project queries natively via the projectPath parameter, and Claude uses it automatically when you reference another codegraph-enabled project. Thanks for pushing in that direction!