DESI MCP Server

A Model Context Protocol (MCP) server for accessing DESI (Dark Energy Spectroscopic Instrument) survey data through dual access methods: Data Lab SQL queries (default) and SPARCL client (backup).

Overview

This MCP server provides fast, comprehensive access to DESI spectroscopic data, enabling AI assistants and researchers to:

• Search DESI objects with coordinate-based queries (nearest, cone, box search)
• Retrieve complete spectral data with wavelength/flux arrays
• Filter by object type (galaxy, quasar, star), redshift, and data release
• Access the full DESI catalog with no row limits via Data Lab SQL
• Cross-survey searches including BOSS and SDSS via SPARCL client

Architecture

Dual Access Methods

Data Lab SQL (Default):

• Fast queries against sparcl.main table
• Efficient Q3C spatial indexing for coordinate searches
• No result limits - access full DESI catalog
• Asynchronous support for large datasets (>100k results)
• Always sorted by distance for accurate "nearest" results

SPARCL Client (Backup):

• Cross-survey access: DESI + BOSS + SDSS
• Box-constraint spatial searches
• Broader survey coverage when needed
• Activated with use_sparcl_client: true

Available Tools

1. search_objects

Primary search interface for DESI astronomical objects

Coordinate Search Modes:

• Nearest Object: ra, dec (finds closest within 0.1°)
• Cone Search: ra, dec, radius (all objects within radius)
• Box Search: ra_min, ra_max, dec_min, dec_max (rectangular region)

Filtering Options:

• object_types: ["GALAXY", "QSO", "STAR"]
• redshift_min/max: Redshift range constraints
• data_releases: Specific releases ["DESI-DR1", "DESI-EDR", "BOSS-DR16"]

Output Control:

• max_results: Limit number of results
• output_file: Save results to JSON
• async_query: Use for large datasets (>100k)
• use_sparcl_client: Switch to SPARCL for cross-survey search

2. get_spectrum_by_id

Retrieve detailed spectrum data by SPARCL UUID

Parameters:

• sparcl_id: SPARCL UUID identifier (required)
• format: "summary" (metadata only) or "full" (complete spectral arrays)

Returns:

• Object metadata (coordinates, redshift, type, survey info)
• Full spectral data (wavelength, flux, model, uncertainties)
• Saves complete data to JSON files for analysis

Installation

1. Install dependencies:

   pip install mcp sparclclient datalab
   

2. Verify access:

   # Test SPARCL
   from sparcl.client import SparclClient
   client = SparclClient()
   
   # Test Data Lab
   from dl import queryClient as qc
   result = qc.query(sql="SELECT COUNT(*) FROM sparcl.main")
   

Usage

Running the Server

python server.py

MCP Client Configuration

{
  "mcpServers": {
    "desi-mcp": {
      "command": "python",
      "args": ["/path/to/server.py"]
    }
  }
}

Example Queries

Find nearest galaxy:

search_objects(ra=9.9443, dec=41.7221, object_types=["GALAXY"])

Cone search for quasars:

search_objects(ra=150.0, dec=2.0, radius=0.1, object_types=["QSO"], redshift_min=2.0)

Box search with cross-survey data:

search_objects(ra_min=150.0, ra_max=151.0, dec_min=2.0, dec_max=3.0, use_sparcl_client=True)

Get complete spectrum:

get_spectrum_by_id(sparcl_id="1270d3c4-9d36-11ee-94ad-525400ad1336", format="full")

Data Coverage

• DESI DR1: ~18+ million spectra
• DESI EDR: ~1.8 million spectra
• Cross-surveys: BOSS DR16, SDSS DR16 (via SPARCL)
• Sky coverage: ~14,000 square degrees
• Wavelength range: 360-980 nm
• Spectral resolution: R ~ 2000-5500

Key Features

✅ Accurate distance sorting: All coordinate searches properly sorted by distance
✅ No result limits: Access complete DESI catalog via SQL
✅ Fast queries: Optimized Q3C spatial indexing
✅ Cross-survey access: DESI + BOSS + SDSS via SPARCL
✅ Complete spectral data: Full wavelength/flux arrays with uncertainties
✅ Async support: Handle large datasets (>100k results)
✅ Error handling: Robust error handling and fallback options

Performance

• Typical coordinate search: < 1 second
• Large dataset queries: Async support for 100k+ results
• Cross-survey searches: 2-5 seconds via SPARCL
• Spectrum retrieval: < 2 seconds for full spectral data

Dependencies

• mcp: Model Context Protocol framework
• sparclclient: DESI SPARCL client for cross-survey access
• datalab: NOIRLab Data Lab query client for SQL access
• Standard library: asyncio, json, logging

License

[Specify license here]

Support

For issues related to:

• DESI data: See DESI collaboration documentation
• SPARCL access: See SPARCL documentation
• MCP framework: See MCP documentation

Version

Current version: 0.1.0 (Phase 1 implementation)