使用说明
ZERO - Universal Data Inspector and AI MCP Auditor for Enterprises
Category: Local data analysis tool for structured/semi-structured files and databases
Brief: Command-line & local web UI to inspect and slice Parquet/CSV/JSON/SQL files without leaving the machine; remote analysts can access it via a secure URL.
Use When
| Scenario | Description | Don’t Use When | Feature | Reason |
|---|---|---|---|---|
| Local Inspection | Inspect Parquet, CSV, JSON, SQL databases without cloud upload or SaaS export | Real-Time Collaboration | Cloud Sync | Does not sync to cloud storage (Dropbox, S3) or send data outside the local machine |
| Restricted Environments | Work over SSH, in air-gapped networks, or on heavily restricted corporate machines | SaaS-Only Data | Pre-built Dashboards | Does not provide pre-built dashboards, charts, or business KPIs (focus is on data inspection) |
| Large Files | Handle files >500 MB where spreadsheet/online tools time out or crash | Shared Dashboards | Auto-Generated Insights | Does not auto-generate summaries or "recommendations" (no LLM business commentary) |
| Privacy-Critical Data | Work with medical, financial, or internal logs that must stay on-prem | Non-Technical Users | Live Streaming | Does not connect to live streaming APIs (Kafka, Pub/Sub) without a local buffer (e.g. local file / DB) |
| Remote Access | Allow data scientists/engineers to inspect data via browser without shell access |
| Category | Formats | Metric | Electron Tools | ZERO |
|---|---|---|---|---|
| Tabular | Parquet, CSV, JSONL, XLSX, ODS | Binary size | 120MB+ | 25MB |
| Document | JSON, XML, TOML, YAML | RAM usage | 2-4GB | <100MB |
| Database | SQLite, PostgreSQL, MySQL | Startup time | 2-5s | <300ms |
| Remote | S3, Kafka, RabbitMQ | Data location | Cloud-first | Local-only |
| Analysis | PII redaction, secret scanning, EXIF metadata |
Architecture
| Component | Description | Technology |
|---|---|---|
| Data Flow | Raw Data → Rust Reader → Node.js MCP Server → AI Client | |
| Rust Layer | Memory-mapped I/O for Parquet/CSV/SQL. Zero-copy processing | Rust |
| MCP Layer | Node.js server exposing tools to AI agents | Node.js |
| Client Layer | Gemini CLI or Claude Desktop integration | AI Agents |
Data Flow Phases
| Phase | Tools | Description |
|---|---|---|
| Perception | list_tree → search_files → scan_patterns |
Map workspace structure and content |
| Analysis | inspect_resource → get_database_schema → (New: Web Project Introspection) |
Calculate quality metrics, visualize schema, and understand web application internals |
| Execution | Apply zero-utils maintenance scripts or ingest operations |
Constraints
| Constraint | Description |
|---|---|
| Read-Only by Default | All tools operate in read-only mode except ingest_data. Prevents accidental data modification |
| Privacy-First | If inspect_resource or scan_patterns returns compromised: true, user action required before proceeding |
| Binary Dependencies | Requires compiled Rust binaries in target/release:- reader (data parsing)- ingestor (SQLite loading) |
| Shell Variability | Commands like find and rg may behave differently across OS platforms |
Remote Access via URL (Key Feature)
| Feature | Description |
|---|---|
| Local Service | ZERO Data Inspector runs as a local service (e.g. Node.js/Python HTTP server) on the data machine |
| Internal Exposure | Internally exposes a simple web UI and REST API on http://localhost:$PORT for file/database inspection |
| Secure Tunnel | Host uses a secure tunnel (e.g. cloudflared, loclx) to expose that local port as a public URL like https://zero-data-abc123.example.com |
| Remote Access | Analysts can open that URL in their browser to: - Browse files in the protected directory - Run ad-hoc queries on Parquet/CSV/JSON/SQL - View samples, schemas, and basic stats |
| Security | Access is secured via: - Short-lived token in the URL (e.g. ?token=xyz) or basic auth- Network-level controls (IP allowlist, firewall rules) |
| Example URL | https://zero-data-abc123.example.com?token=readonly-xyz |
| Note | While remote access provides convenience, performance for very large files (e.g., >500MB) will inherently depend on network bandwidth and latency, and may not be as instantaneous as local access |
Interfaces
| Interface | Description | Technology |
|---|---|---|
| Matrix (TUI) | Terminal interface. For SSH sessions, headless servers, or when GUI overhead matters | Terminal |
| Cortex (GUI) | Tauri-based desktop app. Handles 2GB+ files with native performance | Tauri |
| Eyes (MCP Extension) | Model Context Protocol server for AI agents (Claude, Gemini, Cursor). Provides deep web project introspection, direct schema inspection, and context-aware tool access | AI Agents |
Installation
cargo build --release
./scripts/release-ui.sh # GUI (Tauri)
./scripts/release-tui.sh # Terminal
Binary outputs: ~25MB | RAM: <100MB | Startup: <300ms
MCP Tools (Gemini/Claude Extension)
| Category | Tool | Description | Example Usage |
|---|---|---|---|
| Core Operations | inspect_resource |
Schema + metadata + sample rows from file/URL. Returns information density metrics. | /zero:inspect data.csv 5 |
query_database |
Read-only SQL execution against SQLite/PostgreSQL. | SELECT * FROM users LIMIT 10 |
|
get_database_schema |
Extract table/column structure visualization. | /zero:schema |
|
analyze_join_keys |
Suggest join columns between two data sources. | /zero:join users orders |
|
| Workspace Utilities | search_files |
Regex search via ripgrep. |
/zero:search "error" |
list_tree |
Directory structure up to max depth. | /zero:tree --depth 3 |
|
scan_patterns |
Auto-detect PII, API keys, secrets. | /zero:scan |
|
ingest_data |
Load data into SQLite with optional embeddings. | /zero:ingest data.csv |
|
| Web Project Analysis | analyze_project |
Understand project structure and technologies. | /zero:analyze |
get_components |
Analyze React components and their structure. | /zero:components |
|
get_routing_structure |
Map application routing and paths. | /zero:routes |
|
analyze_dependencies |
Categorize project dependencies. | /zero:deps |
|
get_tailwind_usage |
Analyze Tailwind CSS class usage. | /zero:tailwind |
|
get_hooks_usage |
Detect and quantify React hook usage. | /zero:hooks |
|
analyze_api_calls |
Identify and characterize API calls. | /zero:apis |
|
analyze_database_schema |
Inspect code-defined database schemas. | /zero:db-schema |
Links
License: LICENSE (public domain)
Enterprise Support: Regional deployment consulting available at hi@theneurocatalyst.com
