Run OpenMed's Model Context Protocol (MCP) server so coding agents and chat clients can call clinical NER, PII extraction, and de-identification as tools on-device. It exposes seven specific tools and supports stdio or Streamable HTTP transports for integration into agent workflows.
Tasks
Add an OpenMed MCP server entry to a coding agent config (.mcp.json for Claude Code) using the stdio transport
Start a shared network endpoint by launching the server with the streamable-http transport and proxy it behind your own auth/TLS
Inputs
text
model_name
confidence_threshold
method
lang
Outputs
clinical NER entities
PII/PHI span detections
de-identified text
Limitations and checks
The server has no built-in authentication or TLS for the HTTP transport
Model execution is on-device, requiring adequate local compute and memory
keep_mapping=True generates a re-identification map that must be treated as PHI and never logged
Verify the client spawns the server successfully over stdio using the configured mcpServers JSON entry
Confirm the correct ServiceRuntime environment variables (OPENMED_PROFILE, OPENMED_SERVICE_PRELOAD_MODELS, etc.) are set to avoid first-call latency
Ensure no raw PHI is introduced into logs and that the keep_mapping option is disabled for untrusted agents
AI Search
Find projects, verify facts, compare options, or turn a complex need into an actionable plan
Try a searchA click only fills the search box; you stay in control