DrugBank GraphQL API and MCP Server for Biomedical Data Integration

DrugBank developed a two-layer interoperability stack for cross-platform access to curated pharmaceutical data: a GraphQL API for programmatic querying and a Model Context Protocol (MCP) server for AI-agent integration. The GraphQL layer provides a standards-based interface across five entity types (drugs, proteins, diseases, companies, clinical trials) with ontology-aware search traversing synonym groups and hierarchies. The MCP layer lets compatible AI agents query DrugBank without custom code. Limited release now; GA Q3 2026.

Architecture

Two-layer stack built on open standards:

• Data Layer — GraphQL API: Schema-driven query interface over DrugBank's curated pharmaceutical knowledge base. Supports filtering, pagination, sorting, and aggregation across five entity types.
• AI Integration Layer — MCP Server: Implements the Model Context Protocol (MCP), an open standard for connecting AI models to external data sources. Wraps the GraphQL API with tool definitions, help system, and schema introspection.

Entity Model

Five first-class, interconnected entity types:

• Drugs: Active pharmaceutical ingredients (approved and investigational), with biosimilars and salts grouped into unified entries
• Proteins/Targets: 1:1 mapping to UniProt records, with action types (agonist, inhibitor, substrate) describing drug-protein relationships
• Diseases/Conditions: Organized into synonym groups with preferred terms, placed in a navigable hierarchy aligned to medical specialties
• Companies: Normalized to parent organizations, linked through clinical trial sponsorship
• Clinical Trials: Harmonized and linked across all entity types, with ontology-based matching

Query Capabilities

• Ontology-aware search: Queries traverse synonym groups and disease hierarchies — searching "breast cancer" returns results across all mapped synonyms and subtypes
• Cross-entity traversal: Single query can span drug → target → disease → trial → company relationships
• Aggregation endpoints: Fast counts, distributions, and top-N analyses using the same filter system as paginated queries
• Structured filters: Combinable filter objects with operators (any, not_any, all, exists, empty) and date range support

MCP Server Capabilities

• Tool-based interface: run_query, help, introspect_schema
• Built-in help system with topic-specific guidance for filters, workflows, and best practices
• Schema introspection for dynamic field discovery
• Pagination support for large result sets

Requirements

• Authentication: API key required (issued with DrugBank commercial license)
• Protocol: HTTPS (GraphQL API), stdio or SSE transport (MCP server)
• Query Language: GraphQL (standard specification)
• MCP Compatibility: Any MCP-compatible client (e.g., Claude Desktop, custom AI agents built on Anthropic's Agent SDK, or other MCP-compatible platforms)

Dependencies

• The MCP server requires an active DrugBank API subscription
• MCP clients must support the Model Context Protocol standard (open specification)
• No specific OS requirements — the GraphQL API is cloud-hosted; the MCP server runs as a lightweight process

Data Coverage

• Drugs: approved and investigational compounds with pharmacology, indications, categories
• Proteins: UniProt-mapped targets with drug interaction types
• Diseases: 104,737 synonym groups with hierarchy coverage and cross-ontology codes (MeSH, MedDRA, SNOMED, ICD-10)
• Clinical Trials: harmonized trial data linked to drugs, diseases, and companies
• Companies: normalized sponsor and manufacturer records

Primary Customers

• Pharmaceutical companies: R&D teams performing competitive landscape analysis, target identification, and clinical trial intelligence across drug, disease, and company dimensions
• Systematic review platforms: Tools like DistillerSR that need structured pharmaceutical data for automated literature screening and extraction workflows
• AI agent developers: Teams building biomedical AI assistants that need access to curated drug and disease data without custom API integration
• Academic research groups: Institutions performing drug repurposing research, pharmacological analysis, or clinical trial meta-analyses

Business Partners

• AIDESL consortium partners: The API and MCP server are designed to enable cross-use between consortium collaborators, including DistillerSR (systematic review), and other partners working on AI-driven data extraction from scientific literature
• AI platform providers: Organizations building MCP-compatible tools that could benefit from integrated pharmaceutical data access

Commercial license required. Access to both the GraphQL API and MCP server requires an active DrugBank subscription agreement. Licensing terms are available on request. The MCP server implementation follows the open Model Context Protocol standard, but the underlying DrugBank data and API are proprietary.