code-intelligence — query the code knowledge graph

The codebase-memory MCP server indexes this repo into a queryable graph
(functions, classes, routes, CALLS/DATA_FLOWS/IMPORTS edges, complexity
signals, cross-service links). Reach for these tools before grep/glob for
any structural question — they return resolved symbols and relationships, not
raw text hits.

First move

If nothing is indexed yet (list_projects empty, or you just opened the repo):

index_repository { repo_path: "<repo root>", mode: "moderate" }

Modes: full (all files + similarity/semantic edges), moderate (filtered +
semantic), fast (no semantic), cross-repo-intelligence (match routes/channels
across projects — needs target_projects).

Pick the right tool

Hot-path / bottleneck hunt (Cypher)

Function & Method nodes carry complexity props: complexity (cyclomatic),
cognitive, loop_depth, transitive_loop_depth, linear_scan_in_loop,
alloc_in_loop, recursion_in_loop, unguarded_recursion, recursive.

MATCH (f:Function)
WHERE f.transitive_loop_depth >= 3 OR f.linear_scan_in_loop >= 1
RETURN f.qualified_name, f.transitive_loop_depth, f.linear_scan_in_loop
ORDER BY f.transitive_loop_depth DESC

Workflow tips

• Resolve first, read second. get_code_snippet wants an exact
  qualified_name — get it from search_graph, don't guess.
• Paginate. search_graph caps results; check has_more/total and page
  with offset for broad queries. Narrow via label, file_pattern,
  min_degree first.
• Impact analysis before edits. trace_path {direction: inbound} on the
  function you're about to change shows every caller you might break.
• Stale graph? Re-run index_repository (or detect_changes to see drift)
  after big edits. aish auto-indexes on repo-open when the gate is on.

The graph replaces the grep→read→grep ping-pong for structural discovery — one
search_graph or trace_path usually answers what three greps would only hint
at.