Delv
CommunityActive· 7d4.3by Christian Hinge

DICOM MCP

Queries, reads, and moves data on DICOM servers such as PACS and VNA, extracting reports and searching patient metadata.

C
Safety & Trust

Delv Safety Grade: C

Score 58/100 · assessed 2026-04-28

Maintainer45
Permissions40
Supply chain65
Transparency70
Incidents100

DICOM MCP is a community server enabling Claude to interact with medical imaging infrastructure (PACS/VNA systems). It's maintained by a solo developer (Christian Hinge) with limited visible activity, creating bus factor concerns for healthcare-critical tooling. The server handles highly sensitive patient data and medical imaging, yet requires network access to hospital systems without clear documentation on authentication, encryption, or HIPAA compliance measures. Supply chain is reasonable via PyPI, but the narrow maintainer base and healthcare context elevate risk. The permissions footprint is substantial: network access to private medical networks, potential PHI exposure, and data movement capabilities. Transparency is adequate with open source code, but lacks security audit trails or compliance documentation. No known incidents, but the combination of solo maintenance and healthcare data handling warrants careful vetting before production use.

Lethal Trifecta (prompt-injection exposure)

ONE OF THREE
Private dataYes
Reads secrets, credentials, private files
Untrusted inputNo
Ingests web pages, PRs, issues, emails
External commsNo
Can send data outbound

Medical imaging is sensitive private data. No untrusted input or outbound by default.

Green flags

  • Open source with visible codebase on GitHub
  • Distributed via standard PyPI package manager
  • Addresses genuine clinical informatics need (DICOM protocol bridge)
  • No known security incidents or CVEs

Red flags

  • Solo maintainer for healthcare-critical infrastructure handling PHI
  • Network access to private medical systems (PACS/VNA) without clear auth docs
  • No visible HIPAA compliance documentation or security audit
  • Data movement capabilities across medical imaging nodes risk PHI leakage
  • Limited maintenance activity for mission-critical healthcare tooling

Permissions requested

Outbound networkPrivate networkIdentity readDB readDB write
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Install

uv tool install dicom-mcp

Review

DICOM MCP bridges Claude to medical imaging infrastructure, specifically PACS (Picture Archiving and Communication Systems) and VNA (Vendor Neutral Archives). If you work in radiology IT, clinical informatics, or build tools for healthcare, this is a rare piece of kit that speaks the DICOM protocol natively from an LLM context. What it actually does: query patient metadata, search for studies by modality or date range, retrieve imaging reports, and move DICOM datasets between nodes. I'd reach for this when prototyping a clinical AI assistant that needs to pull radiology reports or verify whether a patient's CT scan exists in the archive. The server handles C-FIND, C-MOVE, and C-GET operations, which are the bread and butter of DICOM networking. The standout use case is conversational PACS querying. Instead of wrestling with clunky vendor UIs, you can ask Claude to find all chest X-rays for a patient in the last six months, then pull the associated reports. It's particularly useful for audit workflows, where you need to trace imaging studies across multiple systems, or for building custom dashboards that aggregate data from legacy PACS that lack modern APIs. Quirks: you need a DICOM server to point it at, obviously. This isn't a toy you can demo without access to a real or simulated PACS. The repo doesn't include a test server, so if you're not already in a healthcare environment, setup friction is high. Also, DICOM networking is famously finicky. Expect to wrangle AE titles, ports, and firewall rules. The server itself is straightforward, but the ecosystem it plugs into is not. Who shouldn't bother: anyone outside healthcare or medical imaging. If you're not dealing with DICOM datasets daily, this is too niche. Also, if you're looking for image analysis (segmentation, detection), this server doesn't do that. It's purely for metadata and report retrieval, not pixel-level work. Documentation is minimal. The README covers installation but assumes you know your way around DICOM. If you're new to medical imaging protocols, budget time for learning the underlying standards. For those already fluent, though, this is a clean, focused tool that does exactly what it says.
Verdict

Install this if you're building clinical AI tools or need programmatic access to PACS data from Claude. Skip it if you're not in healthcare or don't have a DICOM server to connect to. It's a specialist tool for a specialist audience, but it's the only MCP server doing this job.

Good at

  • Only MCP server that speaks native DICOM protocol, filling a real gap for clinical AI workflows.
  • Handles C-FIND, C-MOVE, and C-GET operations, covering the core PACS query and retrieval tasks.
  • Useful for auditing, report extraction, and building custom interfaces over legacy imaging systems.
  • No environment variables required, keeping the config surface small.

Watch out

  • Requires access to a DICOM server, making it impossible to demo or test without healthcare infrastructure.
  • Minimal documentation assumes you already understand DICOM networking and AE title configuration.
  • DICOM protocol itself is notoriously fiddly, so expect setup friction with ports and firewall rules.
  • No built-in image analysis or pixel-level processing, purely metadata and report retrieval.

Use cases

  • radiology workflows
  • PACS querying
  • medical image retrieval
  • clinical report extraction

Getting started

1. Run `uv tool install dicom-mcp` to install the server. 2. Add it to your Claude Desktop config under `mcpServers`, pointing to the installed binary with no extra environment variables required. 3. Configure your DICOM server details (AE title, host, port) when prompted by the MCP server on first use, or pass them as arguments in the config. 4. Test by asking Claude to query a known patient ID or study instance UID to verify connectivity. 5. Watch out for firewall rules blocking DICOM ports (typically 104 or 11112) and ensure your AE title is registered on the target PACS.

Works with

Claude DesktopClaude CodeCursor

Similar MCPs