JSON MCP Filter

A powerful Model Context Protocol (MCP) server that provides JSON schema generation and filtering tools for local files and remote HTTP/HTTPS endpoints. Built with quicktype for robust TypeScript type generation.

Perfect for: Filtering large JSON files and API responses to extract only relevant data for LLM context, while maintaining type safety.

✨ Key Features

• 🔄 Schema Generation - Convert JSON to TypeScript interfaces using quicktype

• 🎯 Smart Filtering - Extract specific fields with shape-based filtering

• 🌐 Remote Support - Works with HTTP/HTTPS URLs and API endpoints

• 📦 Auto Chunking - Handles large datasets with automatic 400KB chunking

• 🛡️ Size Protection - Built-in 50MB limit with memory safety

• ⚡ MCP Ready - Seamless integration with Claude Desktop and Claude Code

• 🚨 Smart Errors - Clear, actionable error messages with debugging info

Related MCP server: JSON Query MCP

🛠️ Available Tools

json_schema

Generates TypeScript interfaces from JSON data.

Parameters:

• filePath: Local file path or HTTP/HTTPS URL

Example:

// Input JSON
{"name": "John", "age": 30, "city": "New York"}

// Generated TypeScript
export interface GeneratedType {
    name: string;
    age:  number;
    city: string;
}

json_filter

Extracts specific fields using shape-based filtering with automatic chunking for large datasets.

Parameters:

• filePath: Local file path or HTTP/HTTPS URL

• shape: Object defining which fields to extract

• chunkIndex (optional): Chunk index for large datasets (0-based)

Auto-Chunking:

• ≤400KB: Returns all data

• 400KB: Auto-chunks with metadata

json_dry_run

Analyzes data size and provides chunking recommendations before filtering.

Parameters:

• filePath: Local file path or HTTP/HTTPS URL

• shape: Object defining what to analyze

Returns: Size breakdown and chunk recommendations

📋 Usage Examples

Basic Filtering

// Simple field extraction
json_filter({
  filePath: "https://api.example.com/users",
  shape: {"name": true, "email": true}
})

Shape Patterns

// Single field
{"name": true}

// Nested objects
{"user": {"name": true, "email": true}}

// Arrays (applies to each item)
{"users": {"name": true, "age": true}}

// Complex nested
{
  "results": {
    "profile": {"name": true, "location": {"city": true}}
  }
}

Large Dataset Workflow

// 1. Check size first
json_dry_run({filePath: "./large.json", shape: {"users": {"id": true}}})
// → "Recommended chunks: 6"

// 2. Get chunks
json_filter({filePath: "./large.json", shape: {"users": {"id": true}}})
// → Chunk 0 + metadata

json_filter({filePath: "./large.json", shape: {"users": {"id": true}}, chunkIndex: 1})
// → Chunk 1 + metadata

🔒 Security Notice

Remote Data Fetching: This tool fetches data from HTTP/HTTPS URLs. Users are responsible for:

✅ Safe Practices:

• Verify URLs point to legitimate endpoints

• Use trusted, public APIs only

• Respect API rate limits and terms of service

• Review data sources before processing

❌ Maintainers Not Responsible For:

• External URL content

• Privacy implications of remote requests

• Third-party API abuse or violations

💡 Recommendation: Only use trusted, public data sources.

🚀 Quick Start

Option 1: NPX (Recommended)

# No installation required
npx json-mcp-filter@latest

Option 2: Global Install

npm install -g json-mcp-filter@latest
json-mcp-server

Option 3: From Source

git clone <repository-url>
cd json-mcp-filter
npm install
npm run build

⚙️ MCP Integration

Claude Desktop

Add to your configuration file:

{
  "mcpServers": {
    "json-mcp-filter": {
      "command": "npx",
      "args": ["-y", "json-mcp-filter@latest"]
    }
  }
}

Claude Code

# Add via CLI
claude mcp add json-mcp-filter npx -y json-mcp-filter@latest

Or add manually:

• Name: json-mcp-filter

• Command: npx

• Args: ["-y", "json-mcp-filter@latest"]

🔧 Development

Commands

npm run build      # Compile TypeScript
npm run start      # Run compiled server  
npm run inspect    # Debug with MCP inspector
npx tsc --noEmit   # Type check only

Testing

npm run inspect    # Interactive testing interface

📁 Project Structure

src/
├── index.ts                    # Main server + tools
├── strategies/                 # Data ingestion strategies
│   ├── JsonIngestionStrategy.ts  # Abstract interface
│   ├── LocalFileStrategy.ts      # Local file access
│   └── HttpJsonStrategy.ts       # HTTP/HTTPS fetching
├── context/
│   └── JsonIngestionContext.ts   # Strategy management
└── types/
    └── JsonIngestion.ts          # Type definitions

🚨 Error Handling

Comprehensive Coverage

• Local Files: Not found, permissions, invalid JSON

• Remote URLs: Network failures, auth errors (401/403), server errors (500+)

• Content Size: Auto-reject >50MB with clear messages

• Format Detection: Smart detection of HTML/XML with guidance

• Rate Limiting: 429 responses with retry instructions

• Processing: Quicktype errors, shape filtering issues

All errors include actionable debugging information.

⚡ Performance

Processing Times

Size Protection

• 50MB maximum for all sources

• Pre-download checking via Content-Length

• Memory safety prevents OOM errors

• Clear error messages with actual vs. limit sizes

Best Practices

• Use json_dry_run first for large files

• Filter with json_filter before schema generation

• Focus shapes on essential fields only

🌐 Supported Sources

• Public APIs - REST endpoints with JSON responses

• Static Files - JSON files on web servers

• Local Dev - http://localhost during development

• Local Files - File system access

💡 Common Workflows

LLM Integration:

1. API returns large response

2. json_filter extracts relevant fields

3. Process clean data without noise

4. json_schema generates types for safety

Hosted deployment

A hosted deployment is available on Fronteir AI.