poptarts-mcp

Public Model Context Protocol server for DigitalOcean App Platform.

Surfaces

Chaos outcomes

Each chaotic request independently picks one of (~25% each):

1. error_4xx — simulated client error (400 / 401 / 403 / 404 / 429)

2. error_5xx — simulated server error (500 / 502 / 503)

3. delayed — sleep ~10s, then succeed

4. ok — succeed immediately

Override with CHAOS=error_4xx|error_5xx|error|delayed|ok|random (error ≡ error_5xx).

Logging + tracing

Tools emit structured JSON logs: info on success, error on chaotic failure or DO API errors.

Each tool log includes chosen traceId / spanId / traceparent / traceSource when present (no full header dump).

Trace preference order:

1. MCP _meta.traceparent / _meta.tracestate (MCP SDK TRACEPARENT_META_KEY)

2. HTTP traceparent / tracestate

3. Zipkin B3 x-b3-traceid / x-b3-spanid

Related MCP server: Jarvis MCP Server

Local development

export DO_API_TOKEN=dop_v1_...
npm install
npm run dev
# MCP:      http://127.0.0.1:8080/mcp
# Droplet:  http://127.0.0.1:8080/droplet/123456789
# Droplets: http://127.0.0.1:8080/droplets
# Flavor:   http://127.0.0.1:8080/flavor
# Valkey:   curl -X POST http://127.0.0.1:8080/valkey/cpu-burn -H 'content-type: application/json' -d '{}'
# Health:   http://127.0.0.1:8080/health

Environment:

Valkey CPU burn

Each POST /valkey/cpu-burn (or MCP start_valkey_cpu_burn) starts one long-lived job: a dedicated Valkey connection that loops a busy Lua EVAL for 5 minutes. Jobs stack — fire hundreds/thousands of requests to keep the Valkey command thread pegged. Every request is logged with its parameters. POST /valkey/cpu-burn/stop aborts all jobs.

# Stack many 5-minute jobs
for i in $(seq 1 200); do
  curl -sS -X POST https://<app>/valkey/cpu-burn \
    -H 'content-type: application/json' -d '{}' &
done
wait

Deploy to DigitalOcean App Platform

1. Create the app from aito-do/poptarts-mcp (Dockerfile, port 8080, health /health).

2. Set secret DO_API_TOKEN in App Platform env (encrypted).

3. Or: doctl apps create --spec .do/app.yaml then set the secret in the UI / doctl.

doctl apps create --spec .do/app.yaml

Connect an MCP client

https://<your-app>.ondigitalocean.app/mcp

{
  "mcpServers": {
    "poptarts": {
      "url": "https://<your-app>.ondigitalocean.app/mcp"
    }
  }
}

License

MIT