Code graph (code-review-graph MCP)

One pre-built Tree-sitter knowledge graph over the whole monorepo answers cross-file questions far more cheaply than walking call chains with Grep+Read. It serves four modes — explore, debug, refactor, review — that share the same setup and tools.

Bootstrap once per session

Graph MCP tools are deferred at session start, so a direct call fails with InputValidationError. Before your first graph call, load the schemas with one ToolSearch:

ToolSearch query: "select:mcp__code-review-graph__query_graph_tool,mcp__code-review-graph__get_impact_radius_tool,mcp__code-review-graph__detect_changes_tool,mcp__code-review-graph__get_affected_flows_tool"

Comma-separate whichever tools the task needs. One cheap call unblocks every graph query for the rest of the session.

Keep the graph current

A graph built on another branch makes detect_changes report function names from unrelated files. Verify and rebuild when needed (the PostToolUse hook keeps it in sync after edits, so a manual rebuild is only needed after git checkout / git pull / large merges):

pipx run code-review-graph==2.3.6 status
pipx run code-review-graph==2.3.6 build

Pick your mode

Explore and understand

1. list_graph_stats for overall metrics.
2. list_communities, then get_community for a specific area.
3. semantic_search_nodes to find a function or class by name (keyword match, not natural language).
4. query_graph with callers_of / callees_of / importers_of to trace relationships, or children_of on a class to list its methods.
5. list_flows and get_flow to follow execution paths.

Debug an issue

1. semantic_search_nodes to locate code related to the symptom.
2. query_graph callers_of and callees_of to trace the call chain in both directions — the entry point that triggers the bug is usually upstream.
3. get_flow for full execution paths through suspect areas.
4. get_impact_radius on suspect files to see what else is affected.
5. Recent changes are the most common cause — verify the branch, then use detect_changes.

Plan a refactor

1. get_impact_radius and get_affected_flows to measure blast radius before touching code.
2. refactor_tool mode rename previews every affected location; dead_code finds unreferenced code; suggest proposes community-driven splits.
3. apply_refactor_tool with the refactor_id applies a previewed rename.
4. After changes, detect_changes to verify the impact.

Caveat: refactor_tool, apply_refactor_tool, and find_large_functions are not empirically validated on this codebase. Treat their output as hypotheses, not prescriptions — verify each edit against the real code, and always preview before applying.

Review changes

1. detect_changes for risk-scored change analysis.
2. get_affected_flows for impacted execution paths.
3. get_impact_radius for the blast radius.
4. Test coverage: grep for *.spec.js / *.unit.js (do NOT use tests_for — it returns 0 incorrectly for many files). Suggest specific cases for untested changes.

Report findings grouped by risk (high/medium/low): what changed and why it matters, test-coverage status, suggested improvements, and an overall merge recommendation.

Cross-cutting rules

• Always pass detail_level: "minimal" — standard mode repeats absolute paths per node and inflates tokens ~6x.
• Use fully qualified names: path/to/file.ts::ClassName.methodName. Bare names return an "ambiguous" error.
• Do NOT call get_architecture_overview — it returns ~3.9M characters and overflows context. Use list_communities + get_community instead.
• For single-file structure, grep is cheaper than children_of.
• Full setup, version pin, rebuild, and troubleshooting: .ai/MCP.md ("code-review-graph MCP").

Target: complete any graph task in ≤5 tool calls.