Searchcraft

The Searchcraft MCP Server provides a suite of tools for managing your Searchcraft cluster's Documents, Indexes, Federations, Access Keys, and Analytics. It enables MCP Clients, like Claude Desktop, to be prompted in plain English to perform administrative actions like setting up search indexes, access keys, ingesting documents, viewing analytics, searching indexes, and more.

Building an app in 2 minutes with Searchcraft MCP Server (video link)

Sample Prompts

Here is a sample prompt that could be used once Claude is connected to the Searchcraft MCP Server.

I'd like to create a product search application using the create_vite_app tool.

Please use this JSON dataset https://dummyjson.com/products
First use the Searchcraft create_index_from_json tool to create the index and add the documents.

Then create an API read key for the vite app using the create_key tool.

App details:
- App name: "my-ecommerce-app"
- Endpoint: http://localhost:8000
- Index name: my-ecommerce-app

Available Tools

The Searchcraft MCP Server currently provides three categories of tools, import tools, engine api tools, and app generation tools:

Engine API Tools

These tools provide direct access to your Searchcraft cluster's core functionality for managing indexes, documents, federations, authentication, and search operations.

Index Management

Document Management

Federation Management

Authentication & Key Management

Stopwords Management

Synonyms Management

Search & Analytics

Import Tools

These tools provide workflows for importing JSON data and automatically generating Searchcraft schemas. Perfect for quickly setting up new indexes from existing data sources.

Import Tools Workflow

The import tools are designed to work together in a streamlined workflow:

1. Analyze → Use analyze_json_from_file or analyze_json_from_url to examine your JSON data structure
2. Generate → Use generate_searchcraft_schema to create a customized Searchcraft schema from the analysis
3. Create → Use the Engine API create_index tool to create the index with your generated schema
4. Import → Use add_documents to populate your new index with data

Or use the all-in-one approach:

• One-Step → Use create_index_from_json to analyze, generate schema, create the index, and import all documents in one command

App Generation Tools

These tools create complete, ready-to-run search applications from your JSON data, perfect for prototyping and demos.

App Generation Workflow

The app generation tools provide an end-to-end solution for creating search applications:

1. Data Analysis → Automatically analyzes your JSON structure to understand field types and content
2. Template Generation → Creates optimized search result templates based on your data fields
3. App Creation → Clones and configures a complete Vite + React application
4. Environment Setup → Configures Searchcraft connection settings
5. Ready to Run → Provides a fully functional search app you can immediately start and customize

Testing

The Searchcraft MCP Server includes a comprehensive test suite built with Vitest.

Run Tests

# Run all tests
npm test

# Run tests in watch mode
npm run test:watch

# Run tests with coverage
npm run test:coverage

# Show test reports
npm run test:ui

Test Coverage

• ✅ 89 tests covering core functionality
• ✅ 84%+ coverage on helpers and utilities
• ✅ 93%+ coverage on JSON analyzer
• ✅ 100% coverage on server creation
• ✅ Integration tests for HTTP endpoints
• ✅ Unit tests for all major components

See test/README.md for detailed testing documentation.

Debugging

Claude Desktop Logs

To view Claude Desktop's logs for debugging MCP connections:

yarn claude-logs

Testing with MCP Inspector

The MCP Inspector allows you to test your server tools interactively.

For stdio server (recommended):

yarn inspect

• Choose Transport Type: stdio
• Command: node dist/stdio-server.js

For HTTP server:

yarn start  # Start HTTP server first
yarn inspect

• Choose Transport Type: Streamable HTTP
• URL: http://localhost:3100/mcp

Manual Testing

Test HTTP server:

# Health check
curl http://localhost:3100/health

# Test MCP endpoint
curl -X POST http://localhost:3100/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}'

Test stdio server:

echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}' | node dist/stdio-server.js

Resources

• 📘 Searchcraft Docs
• 🛰️ Vektron Dashboard
• 💬 Searchcraft Discord
• 🧠 Searchcraft Reddit
• 🧪 Searchcraft SDK on npm

Issues and Feature Requests

Visit https://github.com/searchcraft-inc/searchcraft-issues

License

Licensed under the Apache 2.0 License.

Detailed Tool Usage

Using create_index_from_json

The create_index_from_json tool provides a complete workflow to create a Searchcraft index from JSON data in a single command. This is perfect for quickly setting up search indexes from existing datasets. Note, if you know the language of the data you are importing you should specify it with the language parameter (use the ISO 639-1 two letter code for the language)

Parameters

Example Usage

From a URL:

{
  "source": "url",
  "path": "https://api.example.com/products.json",
  "index_name": "products",
  "sample_size": 50,
  "search_fields": ["title", "description", "category"],
  "weight_multipliers": {
    "title": 2.0,
    "description": 1.0,
    "category": 1.5
  }
}

From a local file:

{
  "source": "file",
  "path": "/path/to/data.json",
  "index_name": "my_data",
  "language": "en"
}

What it does

1. Fetches/Reads Data → Downloads from URL or reads from local file
2. Analyzes Structure → Examines JSON to understand field types and patterns
3. Generates Schema → Creates optimized Searchcraft index schema
4. Creates Index → Sets up the index in your Searchcraft cluster
5. Imports Documents → Adds all JSON data as searchable documents
6. Returns Summary → Provides detailed information about what was created

Expected JSON Format

The tool works with various JSON structures:

• Array of objects: [{...}, {...}, ...]
• Object with array property: {"data": [{...}, {...}], "meta": {...}}
• Single object: {...} (will be treated as a single document)

The tool automatically finds the best array of objects to use for the index.

Using create_vite_app

The create_vite_app tool creates a complete, ready-to-run search application from your JSON data. It's perfect for quickly prototyping search interfaces or creating demo applications.

Parameters

Example Usage

If you saw the prompt earlier in the documentation you can easily use the create_vite_app tool with natural language. However, you want finer grain control you can use the tool with JSON parameters.

Creating a product search app:

{
  "data_source": "url",
  "data_path": "https://api.example.com/products.json",
  "app_name": "product-search",
  "VITE_ENDPOINT_URL": "https://your-cluster.searchcraft.io",
  "VITE_INDEX_NAME": "products",
  "VITE_READ_KEY": "your_read_key_here",
  "sample_size": 100,
  "search_fields": ["title", "description", "brand"],
  "weight_multipliers": {
    "title": 2.5,
    "description": 1.0,
    "brand": 1.8
  }
}

Creating a blog search app from local data:

{
  "data_source": "file",
  "data_path": "/path/to/blog-posts.json",
  "app_name": "blog-search",
  "VITE_ENDPOINT_URL": "https://your-cluster.searchcraft.io",
  "VITE_INDEX_NAME": "blog_posts",
  "VITE_READ_KEY": "your_read_key_here"
}

What it does

1. Analyzes Data Structure → Examines your JSON to understand field types and content patterns
2. Generates Search Templates → Creates optimized result display templates based on your data
3. Clones Vite Template → Downloads the official Searchcraft Vite + React template
4. Installs Dependencies → Sets up all required npm packages
5. Configures Environment → Creates .env file with your Searchcraft settings
6. Customizes Templates → Generates dynamic search result components
7. Updates App Code → Modifies the main app with your specific branding and configuration

Generated App Features

The created application includes:

• React + Vite → Modern, fast development setup
• Searchcraft SDK Integration → Full search functionality out of the box
• Responsive Design → Works on desktop and mobile devices
• Auto-generated Templates → Smart result display based on your data structure
• Environment Configuration → Easy setup for different environments
• Development Server → Hot reload for rapid customization

Template Generation Logic

The tool intelligently analyzes your data to create optimal search result templates:

• Title Field Detection → Finds the best field to use as the main title
• Description Field Detection → Identifies descriptive text fields
• Image Field Detection → Locates image URLs for visual results
• Date Field Detection → Finds timestamp fields for temporal sorting
• Additional Fields → Includes other relevant text fields for comprehensive results

Next Steps After Creation

Once the app is created, you can:

1. Start Vite Server:

   Copy & paste — that's itCopy

   cd apps/your-app-name
   yarn dev

2. Customize Styling → Modify CSS and components to match your brand

3. Add Features → Extend with filters, facets, or advanced search options

4. Deploy → Build and deploy to your preferred hosting platform

Prerequisites

• Existing Searchcraft Index → The index specified in VITE_INDEX_NAME should already exist
• Valid Read Key → The VITE_READ_KEY must have read permissions for the index
• Git Available → The tool uses git to clone the template repository
• Node.js & Yarn → Required for dependency installation

Complete Workflow: From JSON to Search App

Here's how to use both tools together to go from raw JSON data to a fully functional search application:

Option 1: Two-Step Process (Recommended for Production)

Step 1: Create the Searchcraft Index

{
  "source": "url",
  "path": "https://api.example.com/products.json",
  "index_name": "products",
  "sample_size": 100,
  "search_fields": ["title", "description", "category", "brand"],
  "weight_multipliers": {
    "title": 2.5,
    "description": 1.0,
    "category": 1.8,
    "brand": 1.5
  },
  "language": "en"
}

Step 2: Create the Search Application

{
  "data_source": "url",
  "data_path": "https://api.example.com/products.json",
  "app_name": "product-search-app",
  "VITE_ENDPOINT_URL": "https://your-cluster.searchcraft.io",
  "VITE_INDEX_NAME": "products",
  "VITE_READ_KEY": "your_read_key_here",
  "sample_size": 100,
  "search_fields": ["title", "description", "category", "brand"],
  "weight_multipliers": {
    "title": 2.5,
    "description": 1.0,
    "category": 1.8,
    "brand": 1.5
  }
}

Option 2: App-Only Process (For Existing Indexes)

If you already have a Searchcraft index set up, you can jump straight to creating the app:

{
  "data_source": "url",
  "data_path": "https://api.example.com/products.json",
  "app_name": "my-search-app",
  "VITE_ENDPOINT_URL": "https://your-cluster.searchcraft.io",
  "VITE_INDEX_NAME": "existing_index",
  "VITE_READ_KEY": "your_read_key_here"
}

Benefits of the Two-Step Approach

• Index Optimization → Fine-tune your search index separately from the UI
• Multiple Apps → Create different search interfaces for the same data
• Production Ready → Better separation of concerns for production deployments
• Easier Debugging → Test search functionality independently of the UI

Getting Started

Environment Variables

Create .env file at the project's root and fill in the values:

# Server Config
USER_AGENT=searchcraft-mcp-server/<project-version>
DEBUG=true
PORT=3100

# Searchcraft Config
ENDPOINT_URL= # The endpoint url of your Searchcraft Cluster
CORE_API_KEY= # The Searchcraft API key of your Searchcraft cluster. Must match the permissions required by the tools you are using.

.env sample

Remote Usage

If you have already created an index through Vektron on Searchcraft Cloud you may use the write key for the index you are trying to access and use the MCP server for API operations that don't require admin privileges. IMPORTANT: If you use the MCP server with a write key it should NOT be publicly exposed to the internet. Write keys are intended to be secured and by running an MCP server any user with access to the MCP server will be able to write to the index or delete data.

Installation & Setup

Make sure your environment has the correct version of node selected.

nvm use

Install dependencies with yarn

yarn

Build the server

yarn build

This creates two server versions:

• dist/server.js - HTTP server for testing and remote deployment
• dist/stdio-server.js - stdio server for Claude Desktop

Usage

Option 1: Claude Desktop (stdio) - Recommended

For local use with Claude Desktop, use the stdio version which provides better performance and reliability.

claude_desktop_config.json

{
  "mcpServers": {
    "searchcraft": {
      "command": "node",
      "args": [
        "/path/to/searchcraft-mcp-server/dist/stdio-server.js"
      ]
    }
  }
}

The claude desktop config file can be found at:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

If the file doesn't exist, create it.

Option 2: Claude Code

For use with Claude Code, use the CLI to configure the MCP server:

Basic setup:

# Add the Searchcraft MCP server to Claude Code
claude mcp add searchcraft -- node /path/to/searchcraft-mcp-server/dist/stdio-server.js

With environment variables:

# Add with your Searchcraft cluster configuration
claude mcp add searchcraft \
  --env ENDPOINT_URL=https://your-cluster.searchcraft.io \
  --env CORE_API_KEY=YOUR_API_KEY \
  -- node /path/to/searchcraft-mcp-server/dist/stdio-server.js

Configuration scopes:

• --scope local (default): Available only to you in the current project
• --scope project: Shared with team via .mcp.json file (recommended for teams)
• --scope user: Available to you across all projects

Managing servers:

# List configured servers
claude mcp list

# Check server status
/mcp

# Remove server
claude mcp remove searchcraft

Option 3: Open WebUI (via Pipelines)

Open WebUI supports MCP servers through its Pipelines framework. This requires creating a custom pipeline that bridges your MCP server to Open WebUI.

Step 1: Start the Searchcraft MCP HTTP server

yarn start  # Starts HTTP server on port 3100

Step 2: Create an MCP Pipeline for Open WebUI

Create a file called searchcraft_mcp_pipeline.py:

"""
title: Searchcraft MCP Pipeline
author: Searchcraft Team
version: 1.0.0
license: Apache-2.0
description: A pipeline that integrates Searchcraft MCP server with Open WebUI
requirements: requests
"""

import requests
import json
from typing import List, Union, Generator, Iterator
from pydantic import BaseModel


class Pipeline:
    class Valves(BaseModel):
        MCP_SERVER_URL: str = "http://localhost:3100/mcp"
        ENDPOINT_URL: str = ""
        CORE_API_KEY: str = ""

    def __init__(self):
        self.name = "Searchcraft MCP Pipeline"
        self.valves = self.Valves()

    async def on_startup(self):
        print(f"on_startup:{__name__}")

    async def on_shutdown(self):
        print(f"on_shutdown:{__name__}")

    def pipe(
        self, user_message: str, model_id: str, messages: List[dict], body: dict
    ) -> Union[str, Generator, Iterator]:
        # This pipeline acts as a bridge between Open WebUI and your MCP server
        # You can customize this to handle specific Searchcraft operations

        # Example: If user mentions search operations, route to MCP server
        if any(keyword in user_message.lower() for keyword in ['search', 'index', 'document', 'searchcraft']):
            try:
                # Initialize MCP session
                init_payload = {
                    "jsonrpc": "2.0",
                    "id": 1,
                    "method": "initialize",
                    "params": {
                        "protocolVersion": "2025-06-18",
                        "capabilities": {},
                        "clientInfo": {"name": "open-webui-pipeline", "version": "1.0.0"}
                    }
                }

                response = requests.post(self.valves.MCP_SERVER_URL, json=init_payload)

                if response.status_code == 200:
                    # Add context about available Searchcraft tools
                    enhanced_message = f"""
{user_message}

[Available Searchcraft MCP Tools: create_index, delete_index, add_documents, get_search_results, list_all_indexes, get_index_stats, create_key, delete_key, and 20+ more tools for managing Searchcraft clusters]
"""
                    return enhanced_message

            except Exception as e:
                print(f"MCP connection error: {e}")

        return user_message

Step 3: Install the Pipeline in Open WebUI

1. Via Admin Panel:

   • Go to Admin Settings → Pipelines
   • Click "Add Pipeline"
   • Paste the pipeline code above
   • Configure the valves with your Searchcraft settings:
     • MCP_SERVER_URL: http://localhost:3100/mcp
     • ENDPOINT_URL: Your Searchcraft cluster URL
     • CORE_API_KEY: Your Searchcraft API key

2. Via Docker Environment:

   Copy & paste — that's itCopy

   # Save the pipeline to a file and mount it
   docker run -d -p 3000:8080 \
     -v open-webui:/app/backend/data \
     -v ./searchcraft_mcp_pipeline.py:/app/backend/data/pipelines/searchcraft_mcp_pipeline.py \
     --name open-webui \
     ghcr.io/open-webui/open-webui:main

Step 4: Configure Open WebUI to use Pipelines

1. Start Open WebUI with Pipelines support:

   Copy & paste — that's itCopy

   # Using Docker Compose (recommended)
   services:
     openwebui:
       image: ghcr.io/open-webui/open-webui:main
       ports:
         - "3000:8080"
       volumes:
         - open-webui:/app/backend/data
       environment:
         - OPENAI_API_BASE_URL=http://pipelines:9099
         - OPENAI_API_KEY=0p3n-w3bu!
   
     pipelines:
       image: ghcr.io/open-webui/pipelines:main
       volumes:
         - pipelines:/app/pipelines
       environment:
         - PIPELINES_API_KEY=0p3n-w3bu!

2. In Open WebUI Settings → Connections:

   • Set OpenAI API URL to your Pipelines instance
   • Enable the Searchcraft MCP Pipeline

Option 4: HTTP Server (for testing/remote deployment)

Start the HTTP server for testing, debugging, or remote deployment:

yarn start  # Starts HTTP server on port 3100

For Claude Desktop with HTTP server, you'll need mcp-remote:

claude_desktop_config.json

{
  "mcpServers": {
    "searchcraft": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://localhost:3100/mcp"
      ]
    }
  }
}

Option 5: Docker

Run the Searchcraft MCP server in a Docker container for easy deployment and portability.

Build the Docker image:

docker build --load -t searchcraft-mcp-server .

Run the container:

docker run -it -p 8000:8000 \
  --name searchcraft-mcp-server \
  -e ENDPOINT_URL="https://your-cluster.searchcraft.io" \
  -e CORE_API_KEY="your_searchcraft_core_API_key" \
  searchcraft-mcp-server

Test the server:

# Health check
curl http://localhost:8000/health

# Test MCP endpoint
curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}'

Remote inspection with MCP Inspector:

npx @modelcontextprotocol/inspector --transport http --server-url http://localhost:8000/mcp

Docker Configuration:

• Uses Node.js 22-slim as the base image
• Exposes port 3100 by default (configurable via PORT environment variable)
• Automatically handles graceful shutdown on SIGINT/SIGTERM
• Optimized for production with minimal image size

Environment Variables:

• PORT - HTTP server port (default: 8000)
• ENDPOINT_URL - Your Searchcraft cluster endpoint URL
• CORE_API_KEY - Your Searchcraft API key
• DEBUG - Enable debug logging (optional)

Available Scripts

# Development
yarn dev          # Watch HTTP server
yarn dev:stdio    # Watch stdio server

# Production
yarn start        # Start HTTP server
yarn start:stdio  # Start stdio server

# Testing
yarn inspect      # Launch MCP inspector
yarn claude-logs  # View Claude Desktop logs

Deployment Options Comparison

Use stdio when:

• Using Claude Desktop or Claude Code locally
• You want the absolute best performance
• You prefer direct process communication

Use HTTP (local) when:

• You need to test/debug the HTTP interface
• You're developing custom integrations
• You need multiple concurrent local clients

Use Docker when:

• You need remote deployment
• You want easy, reproducible setup
• You're deploying to cloud platforms
• You want isolation and security
• You need to publish your server for inspection