sonic-boom
Encodes a TypeScript/JavaScript codebase into a deterministic PNG heatmap that vision-capable AI models can read to bypass token limits for codebase-scale audits.
README
Sonic Code Sentinel (sonic-boom)
Sonic Code Sentinel encodes an entire TypeScript / JavaScript codebase into a single PNG heatmap that vision-capable AI models can read in one glance β bypassing token limits for codebase-scale audits.
The image is a deterministic function of the AST: same code β byte-identical pixels. Each pixel encodes a category (via color) and severity (via brightness), so the picture itself is the audit summary β no separate lookup required.
π΅ The Spectrogram
The codebase is laid out as a 2D heatmap.
graph LR
Code[Source Code] --> Scanner[AST Scanner]
Scanner --> Map[Deterministic Color Map]
Map --> PNG[Spectrogram PNG]
PNG --> AI[Vision AI Model]
AI --> Fix[Surgical Code Fix]
πΊοΈ Coordinate Mapping
- X-Axis: Each file occupies a contiguous horizontal slot. Within a slot, X is proportional to the line number of the node.
- Y-Axis: Partitioned into four bands (see below). Within the top
ANOMALYband, each Y-row is reserved for one concern β turning the band into a row-per-category heatmap. - Color: Encodes the anomaly category (see catalog below).
- Brightness (Amplitude): Encodes severity β
highis brightest (1.0),medis mid (0.75),lowis dim (0.5). Structural (non-anomaly) pixels fade with nesting depth.
π Frequency Bands
| Hz Range | Layer | Content |
|---|---|---|
| 20 β 8 k | LOGIC |
Source files (.ts, .tsx, .js, .jsx) |
| 8 k β 12 k | STYLES |
Stylesheets (.css, .scss, .less) |
| 12 k β 16 k | MARKUP |
Markup assets (.svg, .html, .md) |
| 16 k β 20 k | ANOMALY |
Per-category rows for detected concerns |
π¬ Detected Anomalies
13 categories. Each has a stable Hz row, a stable color, and a fixed severity β so the heatmap is readable without consulting the mapping table.
| Category | Severity | Detection |
|---|---|---|
| Empty catch block | high | catch (e) {} with empty body |
| Layer Violation | high | File under /components/ importing from /pages/ |
| Massive Component | high | .tsx / .jsx with > 250 lines |
| Prop Overload | high | Interface ending in Props with > 7 members (count surfaced in label) |
Explicit : any type |
med | Parameter, variable, or property typed any |
| Heavy Library Import | med | Imports moment, lodash, or jquery (name surfaced in label) |
| High Complexity | med (high if cc β₯ 20) | Function / method / arrow / class with cyclomatic > 5 or depth > 5 β label includes cc=N, depth=N |
| Heavy Barrel Export | med | index.ts / index.js with > 10 exports |
| Commented-out Code Block | med | Long comment block containing code-like characters |
| Z-Index Escalation | med | z-[N] or z-N where N > 1000 |
| Unresolved TODO/FIXME | med | Leading comment matching `/TODO |
| Tailwind Magic Value | low | Arbitrary -[value] Tailwind class outside var(...) |
| Missing Test File | low | Source file with no sibling *.test.* or *.spec.* |
π¨ Color Legend
| Color | Category |
|---|---|
| White | Layer Violation |
| Bright Red | Empty catch block |
| Light Red | Explicit : any type |
| Purple | Heavy Library Import |
| Orange | Prop Overload |
| Amber | High Complexity (with 5-px horizontal spread for emphasis) |
| Gold | Massive Component |
| Dim Yellow | Heavy Barrel Export |
| Sky Blue | Z-Index Escalation |
| Cyan | Tailwind Magic Value |
| Bright Yellow | Unresolved TODO/FIXME |
| Grey | Commented-out Code Block |
| Dim Blue-Grey | Missing Test File |
Below the ANOMALY band, structural pixels render green / blue / orange for the LOGIC / STYLES / MARKUP layers respectively.
π οΈ MCP Tools
Three tools, used as a pipeline.
| Tool | Purpose |
|---|---|
get_project_spectrogram |
Generates the PNG + mapping table. Returns the PNG (base64), a severity-sorted top-10 anomaly summary, and the color legend. |
resolve_sonic_coordinates |
Translates (x, y) pixel coordinates into { file, line, type, category?, severity?, proximityMatch?, pixelDistance?, context }. |
get_code_snippet |
Fetches a 20-line context window for a given file + line. |
Resolution behavior
- Strict snap: clicks within 12 px of an anomaly pixel resolve to that anomaly, including its canonical category and severity.
- Proximity fallback: a click that lands far from every anomaly in a non-empty file slot still resolves β to the closest in-slot anomaly, flagged with
proximityMatch: trueand apixelDistancein pixels. Replaces the old "Could not resolve coordinates" dead end.
Workflow
- Visualize: agent calls
get_project_spectrogramover the repo root. - Read the heatmap: the vision model identifies a bright pixel cluster by color (e.g. amber row = High Complexity, white row = Layer Violation).
- Resolve:
resolve_sonic_coordinates(x, y)returns file, line, category, severity. - Snippet:
get_code_snippetfor the 20-line window. - Fix: apply patch.
π Getting Started
Prerequisites
- Bun v1.1+
Install
bun install
CLI
Generate a spectrogram for any local directory:
# Scan a project
bun run src/index.ts ../path/to/project
# Exclude additional patterns
bun run src/index.ts ../path/to/project --exclude "**/tests/**"
# Resolve coordinates from a previously generated mapping
bun run src/index.ts fix --coords 150,400 --issue "describe the issue"
Outputs land in ./output/:
spectrogram.pngβ the heatmap.mapping_table.jsonβ per-pixel metadata used by the resolver (file, line, anomaly category, severity).
MCP server
Add to your MCP client config (e.g. Claude Desktop):
{
"mcpServers": {
"sonic-boom": {
"command": "npx",
"args": ["-y", "sonic-boom-mcp@latest"]
}
}
}
The MCP server stores its artifacts under <tmp>/sonic-boom-mcp/; tool calls pass directoryPath to identify the project root.
Inspect MCP locally
bun x npx @modelcontextprotocol/inspector bun run src/mcp-server.ts
Run tests
bun test
The suite covers ignore-file handling, every anomaly detection rule, scan determinism, and resolver round-trip behavior.
π§± Design Properties
- Deterministic: same source β byte-identical PNG and mapping JSON across runs. No RNG anywhere in the scan or render pipeline.
- Heatmap by construction: color = category, brightness = severity, Y-row = concern. The image is the audit, not a decoration.
- Ignore-file aware: respects
.gitignore,.npmignore,.dockerignore,.sonicignore,.eslintignore,.prettierignore, and.gcloudignoreβ both at the repo root and in any subdirectory. Built-in skip list:node_modules,.git,dist,.next,out,build,.vscode,.idea,output.
Sonic Code Sentinel β Hear the code. See the bugs.
π License
This project is licensed under the PolyForm Noncommercial License 1.0.0.
- Free for personal / non-commercial use: research, learning, hobby projects.
- Commercial use requires a separate commercial license β please contact the author for details.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.