Index a repository into a codetopo graph database.

Nodes transitively reachable from a node over calls and imports edges.

Nodes from which a node is transitively reachable over calls and imports edges.

Nodes that break if a node changes: dependents, overriders, and implementers.

Shortest typed path between two nodes, or the reason there is none.

Node and edge totals for an index, with per-kind breakdowns.

Verify the tamper-evident audit chain of an index.

Export the whole graph as a portable snapshot, bound to the audit head.

The diagnostics recorded when the index was built - dependency cycles, multi-definitions, unresolved calls, and the rest - as data, exactly as the CLI prints them.

The top symbols by fan-in, fan-out, blast-radius size, cyclomatic complexity, or a centrality measure (pagerank, betweenness, closeness, eigenvector, katz) - measurements over the graph, with the measurement's method stated in the result. A measure that does not converge on the graph is reported as an absence with its reason, never a partial result. A ranking of measurements, never a score of code quality.

Compare two indexes (or two snapshot files): added, removed, and changed nodes and edges, plus the blast-radius movement of every touched symbol. Give exactly one before source and one after source.

Coupling and cohesion measurements for every type and package (CODE_SCHEMA section 9): CBO, RFC, LCOM pair counts, TCC, ATFD, WMC per type; Ca, Ce, instability, and separately-named external coupling per package. Every metric's method is stated in the result. Measurements only - no thresholds, no verdicts.

The Leiden partition of the dependency graph (Traag et al. 2019; modularity quality, resolution 1.0, fixed published seed) over the undirected CODE_SCHEMA section 9 projection weighted by edge multiplicity: community count, the partition's modularity Q (absent when the graph has no edges), per-community sizes and internal/cut edge counts, and node assignments (limit caps the assignment list; summaries are always full and truncation is stated). Measurements with a stated method - no thresholds, no verdicts.

The safe change order of the graph: topological order of the SCC-condensed CODE_SCHEMA section 9 dependency DAG (Kahn's algorithm over Tarjan components). Groups are emitted in dependency order - every dependency of a group appears in an earlier group - members sorted by id. A multi-node group is a dependency cycle and is flagged cyclic, never disguised as an order.

The graph's layout specification: 'layered' (layout spec v1, the default) or 'force' (Fruchterman-Reingold, layout spec v2 - deterministic, fixed parameters recorded in the spec). Above the force ceilings the answer is a stated decline naming the ceiling, not a partial drawing. The same spec objects serve the CLI's render command and the HTTP render route.

Hotspots (change frequency x complexity, both factors stated) and change coupling (support/confidence as pair counts) over a local working tree's git history. History is local-only: without repo_path the answer is the unmeasured report (measured: false with the reason), never invented numbers.

The subgraph within a few hops of a node, in both directions (contains/defines excluded).

One node in full - kind, location, tier, defines and refs - plus its incident edges grouped by kind and direction, with counts (containment excluded). Errors when the id is unknown.

Find symbols by name (case-insensitive substring of the label) - the discovery step before the exact-id queries.

The schema this server speaks: node and edge kinds, tier rules, and the query vocabulary. Needs no database.