Model Context Protocol (MCP) server that lets AI assistants query, read (PDF), and move data on DICOM servers (PACS, VNA, Orthanc, etc.).
https://github.com/ChristianHinge/dicom-mcpEver tried getting an AI assistant to work with medical imaging data? You're stuck writing custom scripts to query PACS servers, extract report text from DICOM files, and move studies between systems. It's a mess of DICOM protocol handling, PDF parsing, and brittle integration code.
This MCP server changes that. Your AI assistant can now directly query DICOM servers, read embedded PDF reports, and orchestrate data movement across medical imaging systems—all through natural conversation.
Healthcare developers know the pain: DICOM servers speak their own protocol, medical imaging workflows are complex, and AI assistants can't natively interact with any of it. You end up building custom APIs, maintaining brittle scripts, or manually extracting data for analysis.
dicom-mcp bridges this gap directly. Instead of writing query logic, your AI assistant can ask "What CT scans does John Doe have from last month?" and get structured results. Instead of manually extracting report text, it can pull PDF content from DICOM instances automatically.
Here's what that looks like in practice:
User: "Any significant findings in John Doe's previous CT report?"
AI → query_patients → query_studies → query_series → extract_pdf_text_from_dicom
AI: "The report from 2025-03-26 mentions a history of splenomegaly"
User: "Send his latest scan to the segmentation server for volume analysis"
AI → query_studies → move_series (to segmentation node) → extract_pdf_text_from_dicom
AI: "Scan sent to segmentation server, volume report shows spleen at 350cm³"
Query Complex Metadata Without Writing SQL: Ask for "all MRI studies from Q4 2024 with contrast" instead of crafting DICOM C-FIND queries with proper attribute tags.
Extract Report Text Automatically: Pull structured text from embedded PDF reports in DICOM instances. No more manual downloading and parsing.
Orchestrate Multi-System Workflows: Chain operations across different DICOM nodes—query from your main PACS, send to AI processing servers, retrieve results, all in one conversation.
Rapid Research Data Analysis: "Find all chest CT studies with pneumonia mentions in the last 6 months" becomes a simple request instead of complex database queries.
The server handles the DICOM protocol complexity while exposing clean, chainable tools:
Each tool returns structured data that your AI can reason about and chain into complex workflows. You're not limited to simple queries—build sophisticated medical imaging pipelines through conversation.
Drop it into your existing MCP setup with a simple config file:
nodes:
main:
host: "your-pacs.hospital.local"
port: 4242
ae_title: "MAIN_PACS"
ai_processing:
host: "ai-server.local"
port: 11112
ae_title: "AI_NODE"
Add to your Claude Desktop or other MCP client:
{
"mcpServers": {
"dicom": {
"command": "uv",
"args": ["tool", "dicom-mcp", "/path/to/config.yaml"]
}
}
}
Your AI assistant immediately gains DICOM capabilities. No API development, no protocol handling, no PDF parsing libraries.
Medical imaging generates massive amounts of data that's locked away in proprietary systems. Radiologists write detailed reports that live in PDF files embedded in DICOM objects. Research requires querying across years of studies with complex criteria.
Traditional approaches require specialized DICOM knowledge, custom integration work, and maintaining separate systems for data access and analysis. This server eliminates that friction—your AI becomes a natural interface to medical imaging infrastructure.
For research workflows, data analysis, and clinical decision support systems, you now have a direct path from natural language questions to medical imaging data and insights.
Perfect for healthcare software developers building on medical imaging systems, research teams analyzing large datasets, and anyone tired of writing custom DICOM integration code.