Agent / MCP server¶

pydcm.mcp is an in-process Model Context Protocol server: it exposes pydcm's live-object capabilities — the ones that work on DICOM objects in memory, not just files — to any MCP-aware agent runtime, over stdio.

Run it¶

$ python -m pydcm.mcp          # MCP stdio server; -v logs JSON-RPC to stderr

Or embed it in your own process:

import pydcm.mcp
pydcm.mcp.serve(verbose=True)   # serves over stdin/stdout until EOF

Connect an MCP client¶

Point any MCP client (Claude Desktop, Claude Code, your own runtime) at the module entry point:

{
  "mcpServers": {
    "pydcm": {
      "command": "python",
      "args": ["-m", "pydcm.mcp"]
    }
  }
}

Use the interpreter that has pydcm installed (e.g. the absolute path to a venv's python). Restart the client; the tools below appear, callable by name with schema-checked arguments.

Tools¶

The server registers the capabilities that benefit from running inside Python — structured-object reading, volume/geometry work, transforms, and the EHR bridges:

Tool	Does
`dicom_content`	Semantic JSON of a structured object (SEG / RT Struct / RT Plan / RT Dose / PS / Waveform)
`rtdose_info`	RT Dose grid summary (units, scaling, geometry, stored DVHs)
`volume_info`	Assemble a series → 3-D volume; report shape / spacing / affine
`volume_to_nifti`	Series → NIfTI on disk (validated affine)
`resample_volume_to_spacing`	Resample a series to a target mm spacing → NIfTI
`bids_sidecar`	BIDS JSON sidecar (acquisition parameters) from a series
`dicom_to_fhir`	DICOM instance → FHIR R4 `ImagingStudy`
`hl7_parse` / `hl7_build_oru`	Parse an HL7 v2 message / build an `ORU^R01` result

Each tool advertises a JSON Schema (inputSchema) over tools/list, so the agent gets argument validation for free.

When to use it¶

pydcm.mcp runs inside your Python process, so it hands back live results — an assembled volume's geometry, a resampled NIfTI on disk, a parsed RT Dose grid — the things that are Python objects, not just files. Reach for it when the agent is already working in Python and you want it to drive pydcm's volume / transform / RT / EHR capabilities directly.