deploying-openmed-mcp

On-device clinical NLP execution is achieved by deploying an MCP server that exposes clinical NER, PII extraction, and de-identification tools directly to coding agents over stdio. This ensures sensitive clinical text remains locally controlled without external cloud dependencies.

Agent integration for clinical NLP requires configuring compatible MCP clients to launch the server via local stdio transport. This setup exposes seven clinical NLP tools, enabling coding agents to execute de-identification and NER tasks directly.

Shared team endpoints for clinical NLP are supported by deploying the MCP server using Streamable HTTP transport. This configuration requires appropriate authentication, TLS controls, and HTTP security boundaries to regulate network access.

De-identification and PII extraction are clinical NLP processes that identify and remove protected health information from clinical text. Exposing these capabilities as MCP tools allows agents to produce de-identified synthetic text locally.

Model lifecycle management, including model discovery, preloading, memory limits, and keep-alive behavior, is configurable for production operations. These settings optimize the performance of on-device clinical NLP deployments within containerized environments.

Local stdio execution limits MCP agent integration to individual use, whereas Streamable HTTP supports shared networked deployments. Both transports provide identical clinical NLP tools but require distinct security and ServiceRuntime environment configurations.