MCP-KE

Model Context Protocol (MCP) server providing cosmology analysis tools for LLM agents.

What is MCP-KE?

An MCP server that exposes 20+ tools for cosmological analysis:

• Domain Tools: Load data, compute power spectra, create visualizations
• Agent Tools: AI-powered workflows for complex multi-step analysis

Documentation

• DESIGN.md: Architecture, design rationale, and technical details
• CONTRIBUTING.md: Guide for adding new tools

Quick Start

# Install
pip install -e .

# Run server (for testing)
python mcp_server.py

Available Tools

Domain Tools (16 tools)

Data & Models

• load_observational_data - Load eBOSS observational data
• get_lcdm_params, get_nu_mass_params, get_wcdm_params - Cosmology model parameter sets
• create_theory_k_grid - Generate k-value grid for theory predictions

Analysis

• compute_power_spectrum - Compute P(k) for given parameters using CLASS
• compute_all_models - Compute P(k) for all standard models
• compute_suppression_ratios - Calculate P(k)/P_ref(k) ratios

Visualization

• plot_power_spectra - Plot P(k) vs observations
• plot_suppression_ratios - Plot suppression ratios

Helpers

• save_array, load_array, save_dict, load_dict, list_agent_files

Agent Tools (1 tool)

Require OpenAI-compatible LLM API (Anthropic, Google Gemini, OpenAI, etc.)

power_spectrum_agent (DISABLED - requires litellm)

• Commented out in agent_tools/__init__.py
• Use individual tools (compute_power_spectrum, compute_all_models) directly instead
• End-to-end power spectrum analysis workflow
• Orchestrates: data agent → modeling agent → visualization agent
• Example (if re-enabled):

  result = power_spectrum_agent(
      query="Compare ΛCDM and wCDM models with eBOSS data",
      api_key="your-key",
      llm_url="https://api.anthropic.com",
      model_id="claude-3-5-sonnet-20241022"
  )

arxiv_agent

• Search arXiv, download papers, extract and analyze content
• Example:

  result = arxiv_agent(
      query="Find recent papers on galaxy mass estimation methods",
      api_key="your-key",
      llm_url="https://api.anthropic.com",
      model_id="claude-3-5-sonnet-20241022"
  )

Testing

# Unit tests (no API key needed)
pytest

# Integration tests (requires GOOGLE_API_KEY)
export GOOGLE_API_KEY="your-key"
pytest tests/test_agent_tools.py -v

Contributing

See CONTRIBUTING.md

MIT License