Health Automation - Hl7/fhir LLM with Guardrail & Mcp

The workflow JSON

Copy or download the full n8n JSON below. Paste it into a new n8n workflow, add your credentials, activate. Full import guide →

{
  "name": "Health Automation - HL7/FHIR LLM with Guardrail & MCP",
  "nodes": [
    {
      "parameters": {
        "httpMethod": "POST",
        "path": "health-llm-guardrail",
        "responseMode": "responseNode",
        "options": {}
      },
      "id": "webhook",
      "name": "Webhook - Receive HL7/FHIR",
      "type": "n8n-nodes-base.webhook",
      "typeVersion": 1,
      "position": [
        250,
        300
      ]
    },
    {
      "parameters": {
        "mode": "runOnceForAllItems",
        "jsCode": "const hl7 = require('hl7'); // optional if available; manual parsing fallback\n\n// Input data\nconst inputData = $input.first().json.body;\nlet parsed = { type: null, data: null, raw: inputData };\n\n// Detect HL7 (pipe-separated, typical MSH segment)\nif (typeof inputData === 'string' && (inputData.startsWith('MSH') || inputData.includes('\\nMSH'))) {\n  // Simple HL7 to JSON\n  const segments = inputData.split(/\\r?\\n/);\n  const hl7Json = { segments: [] };\n  for (const seg of segments) {\n    const fields = seg.split('|');\n    hl7Json.segments.push({ name: fields[0], fields: fields.slice(1) });\n  }\n  parsed.type = 'HL7';\n  parsed.data = hl7Json;\n  parsed.patientId = hl7Json.segments.find(s => s.name === 'PID')?.fields[2] || 'unknown';\n} else if (typeof inputData === 'object' && (inputData.resourceType || inputData.entry)) {\n  // FHIR resource or bundle\n  parsed.type = 'FHIR';\n  parsed.data = inputData;\n  parsed.patientId = inputData.id || (inputData.subject?.reference) || 'unknown';\n} else {\n  throw new Error('Unsupported format. Please send HL7 string or FHIR JSON.');\n}\n\n// Extract condition/problem for LLM context\nlet clinicalContext = '';\nif (parsed.type === 'HL7') {\n  const obxSegments = parsed.data.segments.filter(s => s.name === 'OBX');\n  clinicalContext = obxSegments.map(s => s.fields[4] || '').join(', ');\n} else if (parsed.type === 'FHIR') {\n  if (parsed.data.resourceType === 'Condition') {\n    clinicalContext = parsed.data.code?.text || parsed.data.code?.coding?.[0]?.display || '';\n  } else if (parsed.data.entry) {\n    const conditionEntry = parsed.data.entry.find(e => e.resource?.resourceType === 'Condition');\n    clinicalContext = conditionEntry?.resource?.code?.text || '';\n  }\n}\n\nreturn {\n  parsed,\n  clinicalContext: clinicalContext || 'No specific condition found.',\n  patientId: parsed.patientId\n};"
      },
      "id": "detect-parse",
      "name": "Detect & Parse HL7/FHIR",
      "type": "n8n-nodes-base.code",
      "typeVersion": 2,
      "position": [
        450,
        300
      ]
    },
    {
      "parameters": {
        "url": "https://your-mcp-server.example.com/context",
        "sendHeaders": true,
        "headerParameters": {
          "parameters": [
            {
              "name": "Content-Type",
              "value": "application/json"
            }
          ]
        },
        "sendBody": true,
        "bodyParameters": {
          "parameters": [
            {
              "name": "patientId",
              "value": "={{ $json.patientId }}"
            },
            {
              "name": "clinicalContext",
              "value": "={{ $json.clinicalContext }}"
            }
          ]
        },
        "options": {}
      },
      "id": "mcp",
      "name": "MCP - Model Context Protocol",
      "type": "n8n-nodes-base.httpRequest",
      "typeVersion": 4,
      "position": [
        650,
        300
      ],
      "notesInFlow": true,
      "notes": "Replace with your actual MCP server endpoint. The server should return relevant clinical guidelines or safety context based on patientId and condition."
    },
    {
      "parameters": {
        "model": "gpt-4",
        "messages": {
          "values": [
            {
              "role": "system",
              "content": "You are a clinical decision support assistant with strict guardrails.\n\nGUARDRAILS (MUST FOLLOW):\n1. NEVER provide definitive diagnoses or treatment plans.\n2. ALWAYS recommend consultation with a licensed physician.\n3. DO NOT predict outcomes or guarantee results.\n4. REFUSE to answer if the input is outside evidence-based guidelines.\n5. If uncertain, respond: 'I cannot provide an answer without verified clinical review.'\n\nYour role is to summarize, flag potential risks, and suggest general educational information only."
            },
            {
              "role": "user",
              "content": "=== PATIENT CLINICAL DATA ===\nHL7/FHIR Extract: {{ $json.clinicalContext }}\nPatient ID: {{ $json.patientId }}\n\n=== MCP CONTEXT (Clinical Guidelines & Safety) ===\n{{ $json.mcpResponse }}\n\nBased on the above, answer the following question (if any was provided in the original request) OR generate a brief risk assessment and educational summary. Ensure your response complies with the guardrails."
            }
          ]
        },
        "options": {
          "temperature": 0.3,
          "maxTokens": 1000
        }
      },
      "id": "llm",
      "name": "LLM with Guardrail Prompt",
      "type": "n8n-nodes-base.openAi",
      "typeVersion": 1,
      "position": [
        850,
        300
      ],
      "credentials": {
        "openAiApi": "<your credential>"
      }
    },
    {
      "parameters": {
        "mode": "runOnceForAllItems",
        "jsCode": "const llmOutput = $input.first().json.message.content;\n\n// Guardrail violation patterns\nconst forbiddenPatterns = [\n  /guarantee/i,\n  /100% cure/i,\n  /will definitely/i,\n  /always safe/i,\n  /no side effects/i,\n  /diagnose you with/i,\n  /prescribe/i,\n  /take this medication/i\n];\n\nlet isSafe = true;\nlet violations = [];\nfor (const pattern of forbiddenPatterns) {\n  if (pattern.test(llmOutput)) {\n    isSafe = false;\n    violations.push(pattern.source);\n  }\n}\n\nlet finalOutput = llmOutput;\nif (!isSafe) {\n  finalOutput = \"[GUARDRAIL ACTIVATED] The generated response contained potentially unsafe content. Please consult a licensed healthcare provider directly.\\n\\nOriginal violations: \" + violations.join(\", \");\n}\n\nreturn {\n  originalLlmResponse: llmOutput,\n  guardrailPassed: isSafe,\n  finalResponse: finalOutput,\n  violationsDetected: violations\n};"
      },
      "id": "guardrail-output",
      "name": "Output Guardrail Check",
      "type": "n8n-nodes-base.code",
      "typeVersion": 2,
      "position": [
        1050,
        300
      ]
    },
    {
      "parameters": {
        "options": {}
      },
      "id": "response",
      "name": "Send Response",
      "type": "n8n-nodes-base.respondToWebhook",
      "typeVersion": 1,
      "position": [
        1250,
        300
      ]
    }
  ],
  "connections": {
    "Webhook - Receive HL7/FHIR": {
      "main": [
        [
          {
            "node": "Detect & Parse HL7/FHIR",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Detect & Parse HL7/FHIR": {
      "main": [
        [
          {
            "node": "MCP - Model Context Protocol",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "MCP - Model Context Protocol": {
      "main": [
        [
          {
            "node": "LLM with Guardrail Prompt",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "LLM with Guardrail Prompt": {
      "main": [
        [
          {
            "node": "Output Guardrail Check",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Output Guardrail Check": {
      "main": [
        [
          {
            "node": "Send Response",
            "type": "main",
            "index": 0
          }
        ]
      ]
    }
  },
  "settings": {
    "executionOrder": "v1"
  },
  "staticData": null,
  "tags": [
    "healthcare",
    "llm",
    "guardrail",
    "hl7",
    "fhir",
    "mcp"
  ],
  "triggerCount": 0,
  "updatedAt": "2025-01-23T12:00:00.000Z",
  "versionId": "1.0.0"
}

Credentials you'll need

Each integration node will prompt for credentials when you import. We strip credential IDs before publishing — you'll add your own.

• openAiApi

About this workflow

Health Automation - HL7/FHIR LLM with Guardrail & MCP. Uses httpRequest, openAi. Webhook trigger; 6 nodes.

Source: https://github.com/mrzasad/n8n-hl7-fhir-llm-guardrail/blob/main/n8n-hl7-fhir-llm-guardrail.json — original creator credit. Request a take-down →

More AI & RAG workflows → · Browse all categories →

Related workflows

Workflows that share integrations, category, or trigger type with this one. All free to copy and import.