Agent Example: Medication Interaction Review

Why This Matters

Traditional drug interaction databases only check prescription-vs-prescription pairs. This scenario requires an agent to correlate data across three provenance boundaries:

Prerequisites

The reference patient pod is included in this repository at ../reference-patient-pod/. It contains 30 days of realistic health data for a fictional patient named Alex Rivera, 52M, Portland, OR.

Option A: MCP via Claude Desktop

Step 1. Add the Cascade MCP server to your Claude Desktop configuration:

{
  "mcpServers": {
    "cascade": {
      "command": "cascade",
      "args": ["serve", "--mcp"],
      "env": {
        "CASCADE_POD_PATH": "/path/to/cascadeprotocol.org/reference-patient-pod"
      }
    }
  }
}

Step 2. Restart Claude Desktop.

Step 3. Open a new conversation and paste the contents of agent-prompt.md as your message.

Step 4. The agent will use the cascade_pod_query and cascade_write MCP tools automatically.

Option B: CLI-as-Skill via Claude Code

Step 1. Open a terminal in this example directory and run Claude Code with the pod path:

export CASCADE_POD_PATH="../reference-patient-pod"
claude

Step 2. Paste the contents of agent-prompt.md as your prompt.

Step 3. The agent will invoke the Cascade CLI directly:

# Step 1: Read medications
cascade pod query ./reference-patient-pod --medications --json

# Step 2: Read supplements
cascade pod query ./reference-patient-pod --supplements --json

# Step 3: Read lab results
cascade pod query ./reference-patient-pod --labs --json

# Step 4: Write back observation
cascade pod write ./reference-patient-pod --type observation --json '{...}'

What the Agent Should Do

1. Read the patient's medication list (both prescribed and self-reported)
2. Identify the ACE inhibitor (lisinopril) + potassium supplement interaction
3. Correlate with the serum potassium lab result (5.1 mEq/L, above the normal range of 3.5–5.0)
4. Write back a flagged observation to the Pod with AIGenerated provenance

The expected output is in expected-output.md. Compare the agent's response against it to assess correctness.

Files in This Example

The following files are included. The Markdown files open as raw text.

Key Cascade Protocol Concepts Demonstrated