crag - Call Graph for AI-Assisted Go Development

AI modifies a function but misses 5 callers that also need updating. crag fixes this — it builds precise call graphs via static analysis, so AI knows exactly what's affected before making changes.

How It Works

crag analyze .          # Build call graph using SSA + VTA static analysis
                        # → Stored in SQLite, fast to query

AI: "Modify ProcessRequest"
 → crag automatically finds all 5 callers
 → AI updates them together, nothing missed

Install

git clone https://github.com/zheng/crag.git
cd crag
go install ./cmd/crag

Quick Start

# 1. Analyze your Go project
crag analyze . -o .crag.db

# 2. Configure MCP for your AI editor (Cursor / Claude Code)
# Add to .cursor/mcp.json or claude_desktop_config.json:

{
  "mcpServers": {
    "crag": {
      "command": "crag",
      "args": ["mcp", "-d", "/absolute/path/.crag.db"]
    }
  }
}

# 3. Keep it updated (pick one)
crag watch . -d .crag.db          # Auto-update on file changes
# or: add `crag analyze . -i` to .git/hooks/post-commit

That's it. Your AI editor can now query call graphs directly.

What AI Can Do With crag

After MCP setup, just ask naturally:

• "Where is HandleRequest called?" → upstream callers
• "If I change BuildSSA, what's affected?" → impact analysis
• "Find all functions containing Auth" → search

CLI Usage

crag impact "HandleRequest" -d .crag.db    # Impact analysis (callers + callees)
crag upstream "db.Query" -d .crag.db       # Who calls this? (recursive)
crag downstream "Process" -d .crag.db      # What does this call?
crag search "Handler" -d .crag.db          # Search functions by name
crag risk -d .crag.db                      # Show high-risk functions
crag implements -d .crag.db                # Interface implementations
crag view -d .crag.db                      # Web UI visualization
crag export -d .crag.db -o crag.md         # Export as Markdown (RAG context)

Why crag?

	Text search (grep)	IDE (gopls)	crag
Interface calls	miss	partial	VTA precise resolution
Persisted & queryable	no	no	SQLite
AI integration	manual copy	no	MCP native
Incremental update	n/a	n/a	Git-aware
Zero CGO	n/a	n/a	Pure Go SQLite

Tech

• Analysis: Go SSA + VTA (Variable Type Analysis) via golang.org/x/tools
• Storage: modernc.org/sqlite (pure Go, single binary)
• CLI: cobra · Web UI: embedded vis.js

License

MIT