AnyRepair

A Rust crate for repairing malformed LLM responses across multiple formats (JSON, YAML, Markdown, XML, TOML, CSV, INI, Diff).

Quick Start

Installation

[dependencies]
anyrepair = "0.1.10"

Basic Usage

use anyrepair::repair;

// Auto-detect format and repair
let malformed = r#"{"name": "John", age: 30,}"#;
let repaired = repair(malformed)?;
println!("{}", repaired); // {"name": "John", "age": 30}

CLI

# Install
cargo install anyrepair

# Auto-detect and repair
anyrepair repair input.json

# Format-specific repair
anyrepair json input.json
anyrepair yaml input.yaml
anyrepair markdown input.md

# Batch process multiple files
anyrepair batch ./data --recursive

# Custom rules management
anyrepair rules add --id "fix_undefined" --format "json" \
  --pattern "undefined" --replacement "null"
anyrepair rules list

# Stream large files
anyrepair stream large_input.json repaired_output.json

# Plugin management
anyrepair plugins list
anyrepair plugins load ./plugins/custom_rules.so

# Validation without repair
anyrepair validate input.json --format json

What's New

v0.1.10 - Latest Release

🔧 8 Format Support

• JSON, YAML, Markdown, XML, TOML, CSV, INI, Diff/Unified Diff
• Auto-detection from malformed content
• Format-specific validation and repair strategies

🐍 Python-Compatible API

• Drop-in compatible with Python's jsonrepair library
• jsonrepair() function and JsonRepair class API
• Easy migration from Python to Rust projects

🔌 MCP Server Integration

• Native Claude Desktop integration via MCP
• 10 MCP tools for format-specific and auto-detect repair
• Simple configuration, zero-dependency setup

📦 Enterprise Features

• Analytics: Track repair metrics, success rates, performance
• Batch Processing: Multi-file processing with format auto-detection
• Validation Rules: Custom rule engine for format validation
• Audit Logging: Compliance logging with event tracking
• Confidence Scoring: Assess repair quality (0.0-1.0 scale)

⚡ Performance & Quality

• 326+ test cases with 100% pass rate
• 99.6% improvement from regex caching
• Streaming support for files larger than RAM
• Parallel strategy application using rayon
• 94% binary size reduction (1.5 MB release builds)

🎯 Plugin System

• Extensible plugin architecture for custom repair strategies
• WASM-based sandboxing support
• Plugin configuration via TOML
• Hot-reload capabilities

See CHANGELOG.md for complete version history.

Why AnyRepair?

LLMs often generate malformed structured data. AnyRepair fixes common issues:

• JSON: Missing quotes, trailing commas, syntax errors
• YAML: Indentation issues, missing colons
• Markdown: Malformed headers, broken links
• XML/TOML/CSV/INI/Diff: Format-specific repairs

Key Features:

• ✅ Auto-detects format from damaged content
• ✅ Multi-format support (8 formats)
• ✅ High performance (regex caching, optimized binaries)
• ✅ MCP server for Claude integration
• ✅ Streaming support for large files
• ✅ 326+ tests, 100% pass rate

Usage Examples

Multi-Format Auto-Detection

use anyrepair::repair;

// JSON - auto-detected and repaired
let json = repair(r#"{"key": value,}"#)?;
// Output: {"key": "value"}

// YAML - auto-detected and repaired
let yaml = repair("name: John\nage: 30")?;

// Markdown - auto-detected and repaired
let markdown = repair("# Header\n[link](url")?;

// Diff - auto-detected and repaired
let diff = repair("@@ -1,3 +1,4 @@\n-line 1\n+line 1 modified")?;

Python-Compatible JSON API

use anyrepair::{jsonrepair, JsonRepair};

// Function-based API (like Python's jsonrepair)
let repaired = jsonrepair(r#"{"name": "John", age: 30,}"#)?;

// Class-based API (like Python's JsonRepair class)
let mut jr = JsonRepair::new();
let repaired1 = jr.jsonrepair(r#"{"key": "value",}"#)?;
let repaired2 = jr.jsonrepair(r#"{name: "John"}"#)?;

Format-Specific Repairers

use anyrepair::{
    JsonRepairer, YamlRepairer, MarkdownRepairer,
    DiffRepairer, XmlRepairer, TomlRepairer
};

// JSON
let mut json_repairer = JsonRepairer::new();
let repaired = json_repairer.repair(malformed_json)?;

// YAML
let mut yaml_repairer = YamlRepairer::new();
let repaired = yaml_repairer.repair(malformed_yaml)?;

// Diff
let mut diff_repairer = DiffRepairer::new();
let repaired = diff_repairer.repair(malformed_diff)?;

// Check if repair is needed
if json_repairer.needs_repair(content) {
    let repaired = json_repairer.repair(content)?;
    let confidence = json_repairer.confidence(&repaired);
}

Streaming Large Files

use anyrepair::StreamingRepair;
use std::fs::File;
use std::io::BufReader;

let input = BufReader::new(File::open("large_file.json")?);
let mut output = File::create("repaired.json")?;

// Configure buffer size (default 8192 bytes)
let processor = StreamingRepair::with_buffer_size(65536);

// Process with automatic format detection
processor.process(input, &mut output, None)?;

// Or specify format explicitly
processor.process(input, &mut output, Some("json"))?;

Batch Processing

use anyrepair::BatchProcessor;

let processor = BatchProcessor::new();

// Process directory with options
let results = processor.process_directory(
    "./data",
    true,  // recursive
    "*.json",  // file filter
)?;

// Get per-file results
for result in results {
    println!("{}: {:?} ({}ms)",
        result.file_path,
        result.status,
        result.repair_time_ms
    );
}

// Get analytics
let analytics = processor.get_analytics();
println!("Success rate: {}%", analytics.success_rate());

MCP Server Integration

The MCP server provides seamless integration with Claude Desktop:

# Install and run MCP server
cargo install anyrepair
anyrepair-mcp

Configure in claude_desktop_config.json:

{
  "mcpServers": {
    "anyrepair": {
      "command": "anyrepair-mcp"
    }
  }
}

Available MCP Tools:

• repair - Auto-detect and repair any format
• repair_json, repair_yaml, repair_markdown, repair_xml
• repair_toml, repair_csv, repair_ini, repair_diff
• validate - Validate content without repair

Usage in Claude:

Please repair this JSON: {"name": "John", age: 30,}
(Claude will use the anyrepair MCP tool to fix it)

See MCP_SERVER.md for complete documentation.

Supported Formats

Advanced Features

Custom Rules

# Add custom repair rule via CLI
anyrepair rules add --id "fix_undefined" --format "json" \
  --pattern "undefined" --replacement "null" --priority 90

# List all rules
anyrepair rules list

# Enable/disable rules
anyrepair rules enable "fix_undefined"
anyrepair rules disable "fix_undefined"

# Remove a rule
anyrepair rules remove "fix_undefined"

Configuration file (anyrepair.toml):

# Custom rules configuration
[[rules]]
id = "fix_trailing_comma"
format = "json"
pattern = ",\\s*}"
replacement = "}"
priority = 95

[[rules]]
id = "fix_js_comments"
format = "json"
pattern = "//.*\\n"
replacement = ""
priority = 80

Plugin System

Extend functionality with custom repair strategies:

use anyrepair::plugin::{Plugin, PluginContext};

struct CustomJsonPlugin;

impl Plugin for CustomJsonPlugin {
    fn name(&self) -> &str { "custom_json" }
    fn version(&self) -> &str { "1.0.0" }
    fn supports_format(&self, format: &str) -> bool {
        format == "json"
    }
    fn execute(&self, content: &str, _ctx: &PluginContext) -> Result<String> {
        // Custom repair logic here
        Ok(content.to_string())
    }
}

Plugin configuration (plugins.toml):

[[plugins]]
name = "custom_json_rules"
path = "./plugins/custom_json_rules.so"
enabled = true
priority = 85
formats = ["json"]

[plugins.config]
strict_mode = true
preserve_comments = false

See Plugin Development Guide for complete documentation.

Enterprise Features

Analytics:

use anyrepair::AnalyticsTracker;

let tracker = AnalyticsTracker::new();
tracker.track_repair_result("json", true, 150.0);
let metrics = tracker.get_metrics();
println!("Success rate: {}", metrics.success_rate());

Validation Rules:

use anyrepair::ValidationRulesEngine;

let engine = ValidationRulesEngine::new();
let violations = engine.validate(content, "json")?;
for violation in violations {
    println!("Rule '{}' violated: {}", violation.rule_id, violation.message);
}

Audit Logging:

use anyrepair::AuditLogger;

let logger = AuditLogger::new();
logger.log_repair("json", input, output, true)?;
logger.log_validation("json", content, &violations)?;

Confidence Scoring:

use anyrepair::ConfidenceScorer;

let scorer = ConfidenceScorer::new();
let confidence = scorer.score(&repaired_content, "json");
if confidence < 0.8 {
    // Manual review may be needed
}

See Enterprise Features for details.

Performance

• Regex Caching: 99.6% performance improvement over uncached operations
• Optimized Binaries: 1.5 MB release builds (94% size reduction)
• Parallel Processing: Multi-threaded strategy application with rayon
• Streaming: Process files larger than available RAM using configurable buffers
• Lazy Evaluation: Skip unnecessary strategies for faster repairs

Build Profiles:

# Standard release (size-optimized)
cargo build --release

# Distribution profile (maximum optimization)
cargo build --profile dist

Testing

• 326+ test cases with 100% pass rate
  • 204 library tests
  • 26 streaming tests
  • 36 fuzz tests
  • 18 complex damage tests
  • 18 complex streaming tests
  • 4 integration tests
  • 2 doctests
• Fuzz testing using proptest for robustness
• Snapshot testing with insta for regression prevention
• Integration tests for end-to-end workflows
• MCP test coverage with 43 comprehensive tests

See TEST_SUMMARY.md for details.

Comparison

Why AnyRepair?

• Most comprehensive format support (8 formats vs JSON-only alternatives)
• Only Rust crate with Python-compatible API and MCP integration
• Enterprise-grade features: analytics, audit logging, validation rules
• Battle-tested with 326+ tests covering real-world LLM failures

Documentation

• ARCHITECTURE.md - System design and architecture
• MCP_SERVER.md - MCP server integration guide
• PLUGIN_DEVELOPMENT.md - Plugin development guide
• TEST_SUMMARY.md - Test coverage details
• CHANGELOG.md - Version history and changes
• INDEX.md - Complete documentation index
• STREAMING_FEATURE.md - Streaming support details
• BUILD_OPTIMIZATION.md - Build optimization guide

Quick Links

• Report Issues: GitHub Issues
• Contributing: See CONTRIBUTING.md (if available)
• Changelog: CHANGELOG.md
• API Docs: docs.rs

Examples

See the examples/ directory for:

• example_plugin.rs - Custom plugin implementation
• mcp_repair_json.rs - MCP JSON repair usage
• mcp_multi_format.rs - Multi-format MCP repair
• mcp_server_usage.rs - MCP server setup and usage

Run examples:

cargo run --example mcp_repair_json
cargo run --example example_plugin

Roadmap

See TODO.md for planned features and improvement areas. Highlights include:

• Additional format support (Properties, .env, GraphQL, SQL, Protobuf)
• CLI enhancements (watch mode, interactive mode, diff preview)
• Web interface and REST API
• Language bindings (Python, Node.js, Go)
• LSP integration and VS Code extension
• Advanced features (format-preserving repairs, LLM-assisted repairs)

License

Apache-2.0

Repository

https://github.com/yingkitw/anyrepair