Summary
index_repository produces a rich graph (nodes + edges + ADRs), but there is no way to visually inspect the result from inside the MCP. After indexing, users have to either spin up Neo4j Browser / Memgraph Lab externally or rely on textual search_graph results.
Proposed behavior
Add an MCP resource (or tool) that exposes a built-in graph viewer — e.g.:
mcp://codebase-memory/graph/{project}/html returning a self-contained interactive page (D3 / Cytoscape / vis-network)- Or a static-file export the user can open directly:
.codebase-memory/graph.html
The viewer should support:
- Click a node → show its qualified name, file, line range, label
- Edge hover → show relation type + source/target
- Filter by label (File / Class / Function / Method / Module …)
- Search box (already covered by
search_graph, but in-graph filtering is faster)
- Zoom / pan / collapse-by-directory
Why
Indexing is the expensive step. Once it's done, exploratory browsing ("what calls this?", "which files touch this module?") is the main use case and works far better visually than Cypher.
Workaround in the meantime
Users can manually decompress .codebase-memory/graph.db.zst and load it into Neo4j Browser / Memgraph Lab — but that's a 5+ min setup that most won't do.
Environment
- Platform: Windows 11
- MCP server version: latest
- Tested on
D-Alpha-evolve-2 (3952 nodes / 6253 edges) — exactly the kind of graph that begs for a viz
Summary
index_repositoryproduces a rich graph (nodes + edges + ADRs), but there is no way to visually inspect the result from inside the MCP. After indexing, users have to either spin up Neo4j Browser / Memgraph Lab externally or rely on textualsearch_graphresults.Proposed behavior
Add an MCP resource (or tool) that exposes a built-in graph viewer — e.g.:
mcp://codebase-memory/graph/{project}/htmlreturning a self-contained interactive page (D3 / Cytoscape / vis-network).codebase-memory/graph.htmlThe viewer should support:
search_graph, but in-graph filtering is faster)Why
Indexing is the expensive step. Once it's done, exploratory browsing ("what calls this?", "which files touch this module?") is the main use case and works far better visually than Cypher.
Workaround in the meantime
Users can manually decompress
.codebase-memory/graph.db.zstand load it into Neo4j Browser / Memgraph Lab — but that's a 5+ min setup that most won't do.Environment
D-Alpha-evolve-2(3952 nodes / 6253 edges) — exactly the kind of graph that begs for a viz