NRTSearch
Exposes Lucene-based search indexes to AI assistants through the NRTSearch MCP server.
NRTSearch MCP Server
Production-ready Model Context Protocol (MCP) server for Lucene/NRTSearch, with first-class support for AI assistants like GitHub Copilot and Claude.
Features
- Exposes NRTSearch/Lucene search as a robust MCP server for AI tools
- Accepts any Lucene query (Boolean, phrase, range, wildcard, fuzzy, etc.)
- Structured logging, retries, and highlight support
- Pure unit-testable search logic with full test coverage
- Easy integration with GitHub Copilot, Claude Desktop, and other MCP clients
- Modern Python packaging and configuration (Pydantic, pyproject.toml)
Quick Start
git clone https://github.com/tvergilio/nrtsearch-mcp-server.git
cd nrtsearch-mcp-server
./quickstart.sh
This will:
- Install all dependencies (including MCP SDK)
- Start the server on the configured port
Usage
CLI / Manual
After installation, you can start the server with either:
# Using the Python module
python -m nrtsearch_mcp.server
# Or, if installed via pip/pipx, use the CLI entrypoint:
nrtsearch-mcp
With GitHub Copilot (VS Code)
- Install VS Code and GitHub Copilot
- Add
nrtsearch-mcpas a Model Context Provider in VS Code settings (see.vscode/settings.json) - Start the server (
./quickstart.shornrtsearch-mcp) - Use Copilot Chat to query your Lucene indexes in natural language
Configuration
The server is configured via environment variables and/or a JSON config file. By default, it looks for:
NRTSEARCH_MCP_CONFIGenv var (path to config)./config.jsonin the current directory~/nrtsearch-mcp-config.jsonin your home directory
Example config:
{
"nrtsearch_connection": {
"host": "localhost",
"port": 8000,
"use_https": false
},
"log_level": "INFO"
}
Key environment variables:
LOG_LEVEL(default: INFO)NRTSEARCH_MCP_CONFIG(optional config path)
API: Search Tool
The main tool is nrtsearch/search:
Parameters:
index(str): Index name (e.g.yelp_reviews_staging)queryText(str): Full Lucene query (e.g.text:(irish AND pub AND (texas OR tx)))topHits(int, default 10): Number of results (1-100)retrieveFields(list, optional): Fields to return (default:["text", "stars"])highlight(bool, optional): Highlight matches
Returns:
- List of hits:
{score, stars, text}
Lucene Query Examples:
text:(irish AND pub AND (texas OR tx))text:"great coffee"stars:[4 TO 5] AND text:(vegan AND brunch)
Testing
Run all tests (unit, no server needed):
pytest -v
Tests cover:
- Success, empty, and multiple hits
- Error handling (HTTP, network, malformed, missing fields)
- Retry logic
- Highlight and custom fields
- Input validation
Project Structure
nrtsearch-mcp-server/
├── nrtsearch_mcp/
│ ├── server.py # Main MCP server and search logic
│ ├── settings.py # Pydantic config
│ └── ...
├── tests/ # Unit tests
├── quickstart.sh # One-step install & run
├── requirements.txt # Python dependencies
├── pyproject.toml # Packaging/metadata
└── ...
License
Apache License 2.0. See LICENSE.
関連サーバー
Jina AI Search
Access Jina AI's Search Foundation APIs for web search, news search, and more, tailored for LLMs.
Sycek OSINT
The Sycek MCP Client is a Model Context Protocol server that gives AI assistants direct access to the Sycek OSINT platform's 20 intelligence tools. Instead of switching between dashboards, you describe what you need and your AI handles the investigation.
grep.app Code Search
Search code across millions of public GitHub repositories using the grep.app API.
regmd-mcp
Medical device regulatory intelligence. Classify devices, query alerts, find pathways across 15 jurisdictions (FDA, EU MDR, HSA, TGA, PMDA, NMPA +9 more). REST API + MCP.
Crossref MCP Server
Search and access academic paper metadata from Crossref.
RateMySupervisor MCP
Query supervisor evaluation data with fuzzy matching for Chinese and Pinyin names.
HR & Compensation MCP Server
H1B salary data, compensation benchmarks, and job market analysis
SearXNG MCP Server
A web search server powered by the SearXNG API.
Perplexity
Interacting with Perplexity
Simple Files Vectorstore
Provides semantic search across local files by creating vector embeddings from watched directories.