Changelog¶

[Unreleased]¶

Preprint Fetch Error: Fixed "Error 400: Could not retrieve publications" when trying to fetch Europe PMC preprints found in search results
Added DOI detection logic to distinguish from PMIDs
Implemented Europe PMC article fetch functionality
Updated routing logic in both unified and individual tools

Dedicated Think Tool: Sequential thinking is now a separate, mandatory tool
think tool MUST be used before any search operations
Provides systematic step-by-step analysis for all biomedical queries
Includes automatic reminders if not used
10 Individual Tools Restored: Direct access tools for specific functionality
Article tools: article_searcher, article_getter
Trial tools: trial_searcher, trial_getter, trial_protocol_getter, trial_references_getter, trial_outcomes_getter, trial_locations_getter
Variant tools: variant_searcher, variant_getter
Enhanced cBioPortal Integration: Now available in individual tools
article_searcher and variant_searcher include include_cbioportal parameter
Centralized helper module for consistent integration
Thinking Usage Tracker: Monitors and encourages proper thinking tool usage
Context-aware tracking across MCP sessions
Automatic reminders in search results when thinking hasn't been used

13 Total Tools: 3 core tools (think, search, fetch) + 10 individual tools
Sequential Thinking: Moved from search(domain="thinking") to dedicated think tool
Search Tool Simplified: Removed thinking domain and parameters from search tool
Query Parameter: Changed from optional (str | None = None) to required with empty default (str = "") for ChatGPT compatibility

MyVariant.info Timeouts: Fixed HTTP client not using configured 120s timeout
Added "myvariant" to aggressive retry domains
Improved error messages for timeout scenarios
Optimized BRAF V600E query pattern
Location Parameter Validation: Added proper validation for trial search location parameters
Ensures lat/long are provided together
Validates distance requires coordinates

Code Organization: Created cbioportal_helper.py to centralize integration logic
Type Safety: Added proper type annotations throughout new modules
Error Handling: Consistent logger usage and error messages

Two Core Tools: search and fetch (following OpenAI MCP guidelines)
Sequential Thinking Integration: Now accessible through search(domain="thinking") instead of a separate tool
Triple-Mode Search: The search tool supports sequential thinking, unified query, and legacy domain-based modes
Backward Compatible: All existing integrations continue to work

Fixed variant fetch error with incorrect parameter names (include_external instead of include_tcga)
Fixed article fetch error where _article_details was called with non-existent fetch_full_text parameter
Fixed article search returning 0 results due to overly restrictive AND logic for keywords
Fixed JSON parsing error in article fetch by calling fetch_articles directly with output_json=True
Fixed trial search "'list' object has no attribute 'get'" error by handling ClinicalTrials.gov API v2 response format
Fixed trial search phase validation error by normalizing phase values (e.g., "Phase 3" → "PHASE3")
Fixed trial search KeyError 'nct_id' by handling nested ClinicalTrials.gov API v2 structure in format_results
Fixed parameter validation errors by accepting both string and list types for search parameters

MCP Tools Consolidation: Consolidated 10 separate MCP tools into 2 unified tools (search and fetch) to align with OpenAI MCP guidelines
Removed individual MCP tool decorators from legacy functions
Legacy functions renamed with leading underscore to indicate internal use
Sequential thinking tool removed from MCP interface (now internal)

Unified Search Tool: Single search tool that handles articles, trials, and variants with domain parameter
Unified Fetch Tool: Single fetch tool that retrieves details for any domain type
OpenAI MCP Compatibility: Search results now follow OpenAI's standardized format with id, title, snippet, url, and metadata fields

CLI interface remains unchanged - all biomcp commands continue to work as before
Internal functions preserved for backward compatibility with CLI

If you were using the MCP tools directly:

Replace article_searcher(...) with search(domain="article", ...)
Replace trial_searcher(...) with search(domain="trial", ...)
Replace variant_searcher(...) with search(domain="variant", ...)
Replace article_details(pmid=X) with fetch(domain="article", id=X)
Replace trial_protocol(nct_id=X) with fetch(domain="trial", id=X, detail="protocol")
Remove usage of sequential_thinking tool (use LLM's internal reasoning instead)