code-review-graph-plus

English | 中文

---

Stop explaining your code to AI over and over.

Every time you start a new chat with Claude, Cursor, or any AI assistant, you spend the first 20 minutes feeding it files, explaining architecture, and correcting misunderstandings.

code-review-graph-plus builds a persistent knowledge graph from your codebase — once. After that, every AI conversation starts with full context: module relationships, call chains, security risks, and architectural hotspots.

No more "let me show you the relevant files." The AI already knows.

---

What it does in 30 seconds

pip install "git+https://github.com/HaipingShi/code-review-graph-plus.git"
cd your-project
code-review-graph-plus install   # hook into Claude / Cursor / Windsurf
code-review-graph-plus build     # one-time scan

Now open any AI chat in that project. Instead of pasting code, ask:

• "Which functions will break if I refactor auth.py?"
• "Find all unprotected paths where passwords reach log sinks."
• "Show me the most critical execution flows in this codebase."
• "Has our architecture cohesion been improving or degrading over the last month?"

The AI answers instantly because it has 32 specialized tools querying a live graph of your code.

---

Before vs After

	Before	After
New AI chat	Paste files, explain architecture, wait for AI to catch up	AI already knows your project structure
Code review	Read diff line-by-line, guess side effects	AI traces impact radius across call chains
Security audit	Manual grep for password, token, log	Automatic source → sink path tracing with risk scoring
Refactoring	Hope you didn't miss anything	AI highlights every affected function and dead code
Architecture health	"Feels messy"	Quantified: cohesion, coupling, community fragmentation trends
Token cost	Burn 5k–20k tokens per chat just explaining context	Zero context tokens — graph is pre-loaded

---

Core capabilities

🔍 Instant project comprehension AI reads your entire codebase structure — files, classes, functions, imports, calls — in a single build. No more file-by-file explanations.

🛡️ Security data-flow audit Automatically finds where sensitive data (passwords, tokens, secrets) is produced, where it might leak (logs, responses, files), and whether it passes through encryption or sanitization along the way.

📈 Technical debt trend tracking Records architecture health snapshots after every build. Track if your code is getting more cohesive or more fragmented over time. Get alerts before it becomes unmanageable.

🏛️ Architecture analysis Detects natural code communities, hotspots (most-connected nodes), architectural chokepoints, and surprising coupling — using only production code, so test files don't pollute the metrics.

---

Architecture

Based on a real build of this project (55 files, 569 nodes, 5,489 edges, 43 communities):

Click the image for an interactive version. Run code-review-graph-plus visualize to generate the same for your project.

---

Quick start

# 1. Install MCP config (auto-detects Claude, Cursor, Windsurf, etc.)
code-review-graph-plus install

# 2. Register your project
cd /path/to/your-project
code-review-graph-plus register .

# 3. Build the knowledge graph (one-time, ~1-3 min)
code-review-graph-plus build

# 4. Check status
code-review-graph-plus status

That's it. Open any AI chat in that project and start asking architecture-level questions.

---

Installation

pip install "git+https://github.com/HaipingShi/code-review-graph-plus.git"

For development:

git clone https://github.com/HaipingShi/code-review-graph-plus.git
cd code-review-graph-plus
pip install -e .

---

What's different from the original

	Original	This fork
Architecture analysis	Includes test code	Production-only — tests excluded from communities, hubs, bridges
Tech debt tracking	None	Snapshots + threshold/trend alerts
Security audit	None	Auto source/sink classification + unprotected path tracing
Community naming	Basic	Enhanced stop-word filtering + cross-community wiki links

---

MCP tools at a glance (32 total)

🏗️ Build & Explore

build_or_update_graph · get_impact_radius · query_graph · semantic_search_nodes · list_graph_stats · traverse_graph

🔍 Review & Changes

get_review_context · detect_changes · get_affected_flows

🏛️ Architecture

list_communities · get_community · get_architecture_overview · list_flows · get_hub_nodes · get_bridge_nodes · get_knowledge_gaps · get_surprising_connections · get_suggested_questions

🔧 Refactoring

refactor · apply_refactor · find_large_functions

📚 Knowledge & Search

embed_graph · generate_wiki · get_wiki_page · get_docs_section · cross_repo_search

📈 Trend Tracking

get_debt_trends · compare_snapshots

🛡️ Security Audit

audit_security_flows · get_security_nodes · get_unprotected_paths · get_security_critical_flows

---

Who is this for?

• Solo developers who switch between AI chats and don't want to re-explain their project every time
• Teams doing code review who want AI to understand cross-file impact before suggesting changes
• Engineers refactoring legacy code who need to know what breaks before they touch anything
• Security-conscious teams who want automated data-flow auditing without manual grep
• Tech leads who want quantified architecture health metrics over time

---

License

MIT. See LICENSE.