# AI in clinical interfaces

Use AI applications and language models in clinical interfaces.

## Get started

Expose a clinical lookup as a tool and an AI agent answers 'did the admission arrive?' end to end.

[Expose services as MCP](https://zato.io/docs/dev/healthcare/ai/agents.html)

```yaml
name: Did it arrive?
actors:
  - id: agent
    label: AI agent
  - id: gateway
    label: MCP gateway
  - id: service
    label: Your lookup service
steps:
  - from: agent
    to: gateway
    label: Did a message arrive for MRN 12345?
    align: left
    items:
      - Claude Code, Cursor or any MCP client
      - The client authenticates with an assigned credential
  - from: gateway
    to: service
    label: The gateway calls patient.message-status
  - from: service
    to: gateway
    kind: return
    label: One received message, control ID MSG00001
  - from: gateway
    to: agent
    kind: return
    label: Received on 2026-03-15 at 12:00
    align: left
    items:
      - Configured response controls run before delivery
      - The enabled audit log records the outcome and message sizes
```

```yaml
name: Message to plain language
actors:
  - id: channel
    label: MLLP channel
  - id: service
    label: Your service
  - id: model
    label: Model, hosted or self-hosted
steps:
  - from: channel
    to: service
    label: An admission arrives as an HL7 message
    align: left
    items:
      - PID|1||12345^^^GH^MR||Smith^John
  - from: service
    to: model
    label: The service sends only the fields the notification needs
    items:
      - The fields remain PHI and use the endpoint selected by the connection
  - from: model
    to: service
    kind: return
    label: Smith was admitted as an inpatient to W3, room 12, at 12:00
  - from: service
    to: channel
    kind: return
    label: The call is recorded
    items:
      - Outcome, duration and token usage
```

## Agents answering your questions

[Expose services as MCP Let AI applications use existing Zato services under the access controls you define.](https://zato.io/docs/dev/healthcare/ai/agents.html) [Interface health Review recorded HL7 failures across interfaces in one result.](https://zato.io/docs/dev/healthcare/ai/interface-health.html) [Feed silence Report interfaces whose latest message exceeds a silence threshold.](https://zato.io/docs/dev/healthcare/ai/feed-silence.html) [Tool schemas Derive the tool definition from a service's docstring and declared fields.](https://zato.io/docs/ai/mcp/tool-schemas.html)

## Your services calling models

[Calling LLMs from clinical services Invoke a model from a clinical service in one line of Python.](https://zato.io/docs/dev/healthcare/ai/llm.html) [Self-hosted models Point the same service at Ollama, vLLM or a corporate model gateway.](https://zato.io/docs/ai/examples/self-hosted-models.html) [Cost and limits Every response reports its token usage and configurable limits cap what an interface may spend.](https://zato.io/docs/ai/cost-and-limits.html)

## Safeguards on patient data

[MCP gateway security Every gateway requires authentication, and separate credentials provide separate client identities.](https://zato.io/docs/ai/mcp/security.html) [Response controls Configure PII detectors, content safeguards and size limits for tool responses.](https://zato.io/docs/ai/mcp/response-controls.html) [Rate limits Attach rate rules to credentials to limit clients independently.](https://zato.io/docs/ai/mcp/rate-limits.html) [MCP audit log Record payload-free metadata for handled MCP requests, including their callers, outcomes and sizes.](https://zato.io/docs/ai/mcp/audit-log.html)