zero-noise
Integrates with OpenAI Codex to provide zero-noise refactoring tools for auditing structural noise, auto-fixing algebraic simplifications, and validating noise score improvements.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@zero-noiseaudit and refactor src/handler.go to eliminate deep nesting"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
zero-noise
Structural silence. Clean code that means what it says.
A universal plugin for Claude Code, Antigravity (agy), OpenCode, and OpenAI Codex (with support for Kimi, Grok, and any MCP-compliant agent) that enforces zero-noise refactoring: DRY, early return, semantic predicates, and algebraic simplification — across Go, Rust, TypeScript, Python, C#, and Zig.
What it does
Audits structural noise: deep nesting (>2 levels), anonymous closures in argument position, raw conditionals, duplicated literals, and commented-out dead code.
Auto-fixes algebraic simplifications with
--fix(De Morgan's laws,endsWithshortcuts, and boolean return cleanups).Proposes machine-applicable fixes via
plan/propose_fixes— guard, predicate, constant, and closure rewrites, each locked to the source SHA-256 (expected_hash) and composable into a single patch.Validates that the noise score is strictly lower after the refactor with compiled native Zig engine with zero external runtime dependencies.
Related MCP server: platao
Install
Quick Install (Universal)
Clone and run the one-stop installer and updater:
git clone https://github.com/xdvi/zero-noise
cd zero-noise
./install.shFlags:
./install.sh(or./install.sh --all): Configures Antigravity, Claude Code, OpenCode, and OpenAI Codex../install.sh --agy: Configures Antigravity only../install.sh --claude: Configures Claude Code only../install.sh --opencode: Configures OpenCode only../install.sh --codex: Configures OpenAI Codex only.
Antigravity (agy)
Install via CLI:
agy plugin install <path-to-zero-noise>Or clone directly into the Antigravity plugins directory:
git clone https://github.com/xdvi/zero-noise ~/.gemini/config/plugins/zero-noiseUpdate:
Re-run agy plugin install <path-to-zero-noise>, or git pull inside ~/.gemini/config/plugins/zero-noise.
Claude Code
Install via marketplace:
# From GitHub repository
claude plugin install zero-noise --marketplace xdvi/zero-noise
# Or from local clone
claude plugin install zero-noise --marketplace ./Update:
claude plugin update zero-noiseOpenCode
Auto-discovered via .opencode/agent/ and .opencode/command/. See .opencode/INSTALL.md for detailed setup.
Install via opencode.json:
Add zero-noise to the plugins array (globally in ~/.config/opencode/opencode.json or project-level opencode.json):
{
"plugins": [
"zero-noise@git+https://github.com/xdvi/zero-noise.git"
]
}Or from a local clone:
{
"plugins": [
"<path-to-zero-noise>"
]
}Global MCP Configuration in opencode.json:
Configure the zero-noise MCP server in ~/.config/opencode/opencode.json (or project-level opencode.json):
{
"mcp": {
"zero-noise": {
"type": "local",
"command": [
"zero-noise",
"mcp"
]
}
}
}Or link automatically via ./install.sh --opencode.
OpenAI Codex
Automatic setup:
./install.sh --codexThis automatically links skills/zero-noise-refactor into ~/.codex/skills/ and registers the MCP server in ~/.codex/config.toml.
Manual configuration (~/.codex/config.toml):
Add the MCP server and marketplace to your Codex configuration:
[mcp_servers.zero-noise]
command = "zero-noise"
args = [
"mcp",
]
[marketplaces.zero-noise]
source = "https://github.com/xdvi/zero-noise.git"
source_type = "git"
[plugins."zero-noise@zero-noise"]
enabled = trueMCP (Model Context Protocol)
For Cursor, Windsurf, Claude Desktop, or custom agent harnesses:
{
"mcpServers": {
"zero-noise": {
"command": "zero-noise",
"args": ["mcp"]
}
}
}Usage
Trigger via slash command:
/zero-noise src/modules/auth/service.goOr ask your agent naturally:
"Refactor this handler, too much nesting"
"Zero-noise refactor this file"
"Clean up this hook, closures everywhere"
"Apply early return and extract predicates here"
The agent audits first, refactors, then re-validates. It never changes behavior — only structure.
CLI Commands
The zero-noise standalone static binary provides high-speed native commands for code analysis and refactoring:
zero-noise audit
Static linter detecting nesting, closures, complex booleans, and commented code blocks:
# Text output
zero-noise audit <path>
# Structured JSON output
zero-noise audit <path> --format=json
# Compact agent-optimized format
zero-noise audit <path> --format=compactzero-noise complexity
Calculates SonarSource Cognitive Complexity per function:
zero-noise complexity <path> --threshold=15zero-noise blast
Calculates definition sites, reference callers, and coupling blast radius:
zero-noise blast <symbol> --root=.zero-noise refactor
Guaranteed safe test-bracketed refactoring runner with automatic rollback on test failure. Dry-run is the default; applying requires explicit --fix:
# Dry run (default)
zero-noise refactor <file>
# Apply algebraic simplifications and verify with test suite
zero-noise refactor <file> --test-cmd="npm test" --fix
# Apply a patch from `plan` / `propose_fixes`
zero-noise refactor <file> --patch-file=fix.patch --patch-kind=guard --expected-hash=<sha256> --test-cmd="npm test" --fixApply contract:
Applications supplying
candidate_content,patch_content, orpatch_filemust declare an explicitpatch_kind(guard,predicate,constant,closure,boolean, ormanual); the engine aborts without one.The engine verifies the declared kind against its own proposals when the patch reproduces one (or a composition), aborting on
classification mismatch; free-form patches keep the declared kind.guardandpredicatepatches requiretest_cmd— the engine refuses them otherwise.expected_hashis the SHA-256 captured at proposal time; any file change aborts withstate lock failed.
zero-noise plan
Lists machine-applicable fix proposals (guard, predicate, constant, closure) for a file or directory — the same report the propose_fixes MCP tool returns. Every proposal carries an expected_hash state lock; proposals never fail the command (exit 0 with findings, 1 on errors):
# Text output
zero-noise plan <path>
# Structured JSON output
zero-noise plan <path> --format=json
# Include the unified diff for each proposal
zero-noise plan <path> --include-patches
# Compose non-overlapping proposals into one applicable patch
zero-noise plan <path> --compose
# Write the composed patch to a file (summary line still printed)
zero-noise plan <path> --compose --out=fix.patch
# Restrict kinds, paginate, or scope the walk
zero-noise plan <path> --kinds=guard,constant --page=1 --page-size=10 --root=.Composition is greedy and deterministic: non-overlapping proposals merge into one candidate, overlapping ones are reported as conflicts (composed: applied=2 conflicts=1). Proposals flagged naming_required carry placeholder names (e.g. isUserReady, extractedClosure1) — propose real domain names to the user before applying.
Git Pre-Commit Hook
Protect your repository by installing the pre-commit hook:
ln -sf ../../scripts/pre-commit .git/hooks/pre-commitConfiguration (.zero-noise.json & .zero-noise-rules.json)
Optional configuration files placed in the repository root:
{
"max_nesting": 2,
"max_closure_len": 80,
"ignore_patterns": [
"node_modules",
"vendor",
".git",
"__pycache__",
"dist",
"build"
]
}Declarative custom rules example (.zero-noise-rules.json):
[
{
"id": "no-direct-fetch",
"pattern": "fetch\\(",
"severity": "warning",
"message": "Use configured apiClient instead of raw fetch()",
"languages": [".ts", ".js"]
}
]Structure
zero-noise/
├── .github/
│ └── workflows/
│ └── ci.yml # GitHub Actions multi-version matrix (Python 3.10-3.13)
├── manifest.json # Single source of truth (DRY)
├── plugin.json # Antigravity root manifest (generated)
├── gemini-extension.json # Gemini / agy metadata (generated)
├── .claude-plugin/
│ └── plugin.json # Claude Code manifest (generated)
├── .opencode/
│ ├── agent/
│ │ └── zero-noise-refactor.md # Symlink -> ../../agents/zero-noise-refactor.md
│ ├── command/
│ │ └── zero-noise.md # Symlink -> ../../commands/zero-noise.md
│ └── INSTALL.md # Quick guide and hooks for OpenCode
├── AGENTS.md # Universal context (OpenCode, agy, Kimi, Grok)
├── README.md # Documentation and install guide
├── install.sh # Universal installer and updater (agy, Claude, OpenCode)
├── LICENSE # MIT
├── commands/
│ └── zero-noise.md # Explicit slash command (/zero-noise)
├── agents/
│ └── zero-noise-refactor.md # Canonical autonomous subagent
├── src/
│ ├── main.zig # CLI entrypoint & router
│ ├── root.zig # Library root module
│ ├── analysis/ # Cognitive complexity, noise, and blast radius
│ ├── mutation/ # Algebraic rewriter and transactional runner
│ ├── storage/ # In-memory state database and state-lock ledger
│ ├── treesitter/ # Tree-sitter C ABI bindings and grammars
│ ├── mcp/ # JSON-RPC 2.0 stdio server & protocol
│ └── cli/ # Subcommand handlers
├── c/ # Tree-sitter C ABI grammars
├── skills/
│ └── zero-noise-refactor/
│ ├── SKILL.md # Agent skill descriptor
│ └── reference/
│ ├── cleanliness_rules.md
│ └── architecture_patterns.md
├── scripts/
│ ├── sync_manifests.py # Manifest synchronizer from manifest.json
│ └── pre-commit # Pre-commit git hook
├── install.sh # Universal POSIX installer
└── tests/
└── fixtures/ # Multi-language AST fixtures (Go, Rust, TS, Py, C#)
├── dirty/
├── clean/
└── fix/Testing & Verification
Run the native test suite:
zig build testLicense
MIT
This server cannot be deployed
Maintenance
Related MCP Connectors
Enforces architectural invariants and code quality rules inside your AI agent's edit loop.
Statically audits MCP tool surfaces for token cost, schema quality, and design issues.
Lets coding agents check their own code for leaked secrets, risky dependencies and AI-code mistakes
Audit and repair any public GitHub repo. Free survey; paid source, unified diffs and a pull request.
Related MCP Servers
- AlicenseAqualityCmaintenanceEnables LLMs to apply Martin Fowler's 71+ refactoring patterns to codebases through a pluggable, language-agnostic architecture. Supports previewing and applying refactorings, analyzing code smells, and inspecting code structure with safe-by-default operations.536 PyPI5MIT
- AlicenseNot gradedqualityBmaintenanceEnables coding agents to run deterministic completeness audits on code before declaring 'done'—catching unwired modules, placebo tests, swallowed errors, and stubs via AST analysis, entirely offline and free.MIT
- FlicenseNot gradedqualityBmaintenanceEnables deterministic AST static analysis and call-graph auditing to identify orphaned functions, unreachable methods, and token-bloat clusters, producing refactoring plans that reduce context waste.7-
- FlicenseNot gradedqualityBmaintenanceEnables coding agents to run deterministic, zero-dependency AST static analysis over source code, building call graphs and dependency metrics to find orphaned functions, unreachable methods, breaking API surface diffs, and token-bloat clusters, then return a structured refactoring plan. It plugs into MCP clients like Claude Desktop, Cursor, and Windsurf via a single Python entrypoint.7-