AlphaGenome MCP Server

# Changelog All notable changes to this project will be documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). ## [Unreleased] ## [0.2.0] - 2025-10-13 ### Added - 10 new Group C wrapper tools (total 20 tools now) - predict_tf_binding_impact: Focus on transcription factor binding effects - predict_chromatin_impact: Assess chromatin accessibility changes - compare_protective_risk: Compare protective vs risk alleles directly - batch_pathogenicity_filter: Filter variants by pathogenicity threshold - compare_variants_same_gene: Rank variants within a single gene - predict_allele_specific_effects: Analyze allele-specific regulatory effects - annotate_regulatory_context: Comprehensive regulatory context annotation - batch_modality_screen: Screen variants across specific regulatory modalities - generate_variant_report: Generate comprehensive clinical report - explain_variant_impact: Human-readable impact explanation ### Changed - README completely restructured to emphasize wrapper architecture - Removed artificial Group A/B/C classification - Organized tools by 7 functional categories - All 20 tools now documented as lightweight wrappers around single predict_variant() API - Updated all query examples to use natural English ("Use alphagenome to...") - Removed parentheses from query examples for better readability - Fixed all README examples to show actual API output (not simplified versions) - Tissue-specific analysis now shows real field names and structure - Variant comparison shows actual output format with splice_delta - Allele comparison shows actual allele_comparisons object structure ### Fixed - batch_modality_screen OutputType serialization error - batch_modality_screen JSON serialization of OutputType enums - Documentation now matches actual API responses (truthfulness principle) ### Documentation - Added paper.tex and paper.pdf with comprehensive technical documentation - Added "Wrapper Versatility Demonstration" section showing same variant analyzed 6 different ways - Enhanced interpretations with percentage explanations - All examples verified with real AlphaGenome API using Alzheimer's disease variants ## [0.1.5] - 2025-10-12 ### Verified - Confirmed full end-to-end functionality with real AlphaGenome API - Tested with rs113700824 variant analysis (APOE region) - Validated Python bridge integration and MCP tool calls - All three tools (predict_variant_effect, analyze_region, batch_score_variants) operational ### Documentation - Verified working installation command: `npx -y @jolab/alphagenome-mcp@latest --api-key YOUR_KEY` - Confirmed compatibility with Claude Desktop MCP integration - Python bridge successfully interfaces with AlphaGenome SDK ## [0.1.4] - 2025-10-12 ### Fixed - Minimized stderr logging to match working MCP servers (sequential-thinking pattern) - Reduced startup message to single line: "AlphaGenome MCP Server running on stdio" - Removed all informational logging during initialization - Removed client initialization logging that could interfere with stdio transport ### Changed - Simplified error messages to avoid redundant stderr output - Followed MCP best practices from @modelcontextprotocol/server-sequential-thinking ## [0.1.3] - 2025-10-12 ### Changed - Optimized server initialization to be fully non-blocking - Reduced logging verbosity in AlphaGenomeClient constructor to avoid stdio interference - Added initialization time warning for first API call (30-60s expected) ### Fixed - MCP health check timeout issues by ensuring no blocking operations during startup - Client initialization now completely deferred until first tool call ## [0.1.2] - 2025-10-12 ### Changed - Removed all "MOCK MODE" warnings from tool descriptions - Updated documentation to reflect real AlphaGenome API usage - Clarified that server uses Google DeepMind's AlphaGenome SDK via Python bridge ## [0.1.1] - 2025-10-11 ### Added - Command-line argument parsing for `--api-key` flag - Support for `claude mcp add` with `--api-key` parameter ### Fixed - Connection failure when using `npx -y @jolab/alphagenome-mcp` without environment variable - API key can now be provided via CLI argument, matching behavior of other MCP servers ## [0.1.0] - 2025-10-11 ### Added - Initial project setup - Real AlphaGenome API integration via Python bridge - Three core MCP tools: predict_variant_effect, analyze_region, batch_score_variants - Comprehensive input validation with Zod - Beautiful Markdown output formatting - TypeScript strict mode configuration - ESLint and Prettier setup - Python bridge for AlphaGenome SDK integration ### Notes - Uses Google DeepMind's AlphaGenome Python SDK - Architecture designed for production use