You can query the NPS Agent for information about U.S. national parks. The NPS Agent is an AI assistant running in the nps-agent namespace that has access to the National Park Service API. It can answer questions about parks, alerts, campgrounds, events, and visitor centers.

How It Works

The NPS Agent runs as a standalone service with its own model and MCP tools. You send questions via HTTP and receive natural language answers. Authentication is handled transparently by the AuthBridge -- you just make the call.

Query the NPS Agent

RESPONSE=$(curl -s --max-time 300 -X POST \
  http://nps-agent.nps-agent.svc.cluster.local:8080/invocations \
  -H "Content-Type: application/json" \
  -d '{"input": [{"role": "user", "content": "Your question about national parks here"}]}')

echo "$RESPONSE" | python3 -c "import sys,json; o=json.load(sys.stdin)['output']; print(next(c['text'] for m in reversed(o) for c in m.get('content',[]) if 'text' in c))"

Important: The NPS Agent may take up to 60 seconds on the first request (cold start). Use --max-time 300 to allow for this.

Input Format

The /invocations endpoint accepts JSON with an input array of messages:

{"input": [{"role": "user", "content": "What national parks are in California?"}]}

Output Format

The response follows the MLflow ResponsesAgent format:

{
  "output": [
    {
      "type": "message",
      "role": "assistant",
      "content": [
        {
          "type": "output_text",
          "text": "California has nine national parks..."
        }
      ]
    }
  ]
}

Extract the answer:

echo "$RESPONSE" | python3 -c "import sys,json; o=json.load(sys.stdin)['output']; print(next(c['text'] for m in reversed(o) for c in m.get('content',[]) if 'text' in c))"

What the NPS Agent Can Answer

The agent has 5 MCP tools connected to the NPS API:

Tool	What It Does	Example Question
`search_parks`	Find parks by state, code, or keyword	"What parks are in Utah?"
`get_park_alerts`	Current alerts and hazards	"Are there any alerts for Yellowstone?"
`get_park_campgrounds`	Campground info and amenities	"What campgrounds are at Grand Canyon?"
`get_park_events`	Upcoming events and activities	"What events are happening at Acadia?"
`get_visitor_centers`	Visitor center locations and hours	"Where are the visitor centers at Zion?"

Examples

Find parks in a state

RESPONSE=$(curl -s --max-time 300 -X POST \
  http://nps-agent.nps-agent.svc.cluster.local:8080/invocations \
  -H "Content-Type: application/json" \
  -d '{"input": [{"role": "user", "content": "What national parks are in Colorado?"}]}')
echo "$RESPONSE" | python3 -c "import sys,json; o=json.load(sys.stdin)['output']; print(next(c['text'] for m in reversed(o) for c in m.get('content',[]) if 'text' in c))"

Check park alerts

RESPONSE=$(curl -s --max-time 300 -X POST \
  http://nps-agent.nps-agent.svc.cluster.local:8080/invocations \
  -H "Content-Type: application/json" \
  -d '{"input": [{"role": "user", "content": "Are there any current alerts or closures at Yellowstone National Park?"}]}')
echo "$RESPONSE" | python3 -c "import sys,json; o=json.load(sys.stdin)['output']; print(next(c['text'] for m in reversed(o) for c in m.get('content',[]) if 'text' in c))"

Get campground info

RESPONSE=$(curl -s --max-time 300 -X POST \
  http://nps-agent.nps-agent.svc.cluster.local:8080/invocations \
  -H "Content-Type: application/json" \
  -d '{"input": [{"role": "user", "content": "What campgrounds are available at the Grand Canyon and what amenities do they have?"}]}')
echo "$RESPONSE" | python3 -c "import sys,json; o=json.load(sys.stdin)['output']; print(next(c['text'] for m in reversed(o) for c in m.get('content',[]) if 'text' in c))"

Health Check

Verify the NPS Agent is running:

curl -s http://nps-agent.nps-agent.svc.cluster.local:8080/ping

Returns 200 OK if healthy.

Run Agent Evaluation

You can trigger an evaluation of the NPS Agent. This runs 6 test cases (parks by state, park codes, campgrounds, alerts, visitor centers) and checks that expected facts appear in the responses.

Trigger an eval run

oc create job nps-eval-$(date +%s) --from=cronjob/nps-eval -n nps-agent

Check eval status

oc get jobs -n nps-agent -l component=eval --sort-by='{.metadata.creationTimestamp}'

Read eval results

JOB_NAME=$(oc get jobs -n nps-agent -l component=eval --sort-by='{.metadata.creationTimestamp}' -o jsonpath='{.items[-1].metadata.name}')
oc logs -l job-name=$JOB_NAME -n nps-agent

The eval output shows pass/fail for each test case, expected facts found, latency per query, and an overall summary. Results are also logged to the NPSAgent experiment in MLflow.

Quick eval (3 test cases only)

To run a faster eval, create the job manually:

oc create job nps-eval-quick -n nps-agent --image=image-registry.openshift-image-registry.svc:5000/nps-agent/nps-agent:latest -- python3 /eval/run_eval.py --quick --standalone

Error Handling

Error	Meaning	Action
Connection refused	NPS Agent pod is down or not deployed	Check `oc get pods -n nps-agent`
Timeout (>300s)	Agent is processing a complex query or cold starting	Retry with a simpler question
Empty response	Agent couldn't find relevant data	Try a more specific query (include park name or state code)
500 error	Agent encountered an internal error	Check NPS Agent logs: `oc logs deployment/nps-agent -n nps-agent`