What You Gain from git-ai Provenance you can trust. Every commit knows whether a human or AI wrote it, which agent, with what confidence, and what human (if any) signed off. The audit trail is cryptographic, not just metadata. Semantic blame. Not "who wrote this line in 2019" but "what intent caused this line to exist, and did an AI write it at 91% confidence with no human review." That's a completely different debugging experience. Policy enforcement at commit time. Rules like "no AI commit touches crypto without a human co-signature" are enforced automatically, not just hoped for in code review. A living semantic index. Your repo accumulates a vector embedding + dependency graph that travels with every clone. AI tools query this instead of re-parsing raw source every time — faster, richer context. AI-native code review. Reviewers see behavioral diffs, not line diffs. "This function now handles null differently" instead of red/green lines.

# 1. Install git-ai pip install git-ai

# 2. Add MCP to Claude Code (one time) claude mcp add git-ai --transport http http://localhost:7420/mcp

# 3. Done — every Claude Code session is now AI-aware