Ground
Code analysis that checks before it claims
AI assistants will tell you files are "95% similar" without actually comparing them. Ground fixes this. It requires your AI to verify before claiming—no more hallucinated duplicates or false positives.
Install in 2 minutes
Copy the config for your AI tool
Add to claude_desktop_config.json
Settings → MCP → View raw config
codex mcp add ground
Works with any MCP client
How it works
Check first
Your AI runs a verification command to actually compare files or count uses
Then claim
Only after checking can it report something as a duplicate, dead code, or orphan
Blocked otherwise
If it tries to claim without checking first, Ground stops it
# First, compare the files
ground compare utils.ts helpers.ts
# Then, make a claim (only works if you've compared)
ground claim duplicate utils.ts helpers.ts "same validation logic"
# Try to claim without checking? Blocked.
✗ Claim blocked
You need to compare these files first:
ground compare utils.ts helpers.tsWhat you can do
20+ tools for finding real problems in your code
Verify
ground_compare— See how similar two files actually areground_count_uses— Find if a function is actually used anywhereground_check_connections— See if a module is connected to your appground_check_environment— Catch Node.js APIs leaking into Workers
Find problems
ground_find_duplicate_functions— Find copy-pasted codeground_find_orphans— Find files nothing importsground_find_dead_exports— Find exports nobody usesground_find_drift— Find where code drifted from your design system
Understand patterns
ground_adoption_ratio— See how consistently you use your tokensground_suggest_pattern— Get suggestions based on your existing codeground_mine_patterns— Discover patterns you're already using
Report findings
ground_claim_duplicate— Report a duplicate (requires verification first)ground_claim_dead_code— Report dead code (requires verification first)ground_claim_orphan— Report an orphan (requires verification first)
Kickstand: 155 scripts became 13
We used Ground to find and consolidate duplicate code in a production codebase. Because every finding was verified, we had zero false positives—every deletion was safe.