ontario-data-mcp: Skill for Claude Code

Generating Smoke Tests

Generate a temporary Python smoke-test script that exercises multiple MCP tools against the live Ontario, Toronto, and Ottawa data APIs, run it, verify all assertions pass, then clean up.

When to Use

• During the release process (after unit tests pass, before version bump)
• After refactoring tool code, server wiring, or lifespan/context changes
• When verifying end-to-end connectivity to data.ontario.ca

Instructions

1. Generate the smoke test script

Write a file called smoke_test.py in the project root. The script must:

1. Discover the latest dataset dynamically — fetch https://data.ontario.ca/feeds/dataset.atom, parse the first <entry> to extract the dataset UUID from its <id> tag (format: https://data.ontario.ca/dataset/{uuid})
2. Set up real dependencies — httpx.AsyncClient, CKANClient, CacheManager with a temp DuckDB path
3. Mock only the FastMCP context — use the pattern from tests/conftest.py:

   from unittest.mock import AsyncMock, MagicMock
   from ontario_data.portals import PORTALS
   ctx = MagicMock()
   ctx.report_progress = AsyncMock()
   ctx.lifespan_context = {
       "http_client": http_client,
       "portal_configs": PORTALS,
       "portal_clients": {},
       "cache": cache,
   }
   

4. Access tools via the FastMCP API: tool = await mcp.get_tool("tool_name") then await tool.fn(...)
5. Exercise this tool chain (each step asserts success before continuing):
   • Ontario: search_datasets(query="ontario", portal="ontario") — assert total_count > 0
   • Ontario: get_dataset_info(dataset_id=<uuid from feed>) — assert returns id or name
   • Ontario: download_resource(resource_id=...) — pick first datastore-active resource; if none found, try the next feed entry (up to 5) until one with a datastore-active resource is found
   • Ontario: query_cached(sql=...) — SELECT COUNT(*) as cnt from the cached table
   • Ontario: cache_info() — assert table_count > 0
   • Toronto: search_datasets(query="toronto", portal="toronto") — assert total_count > 0
   • Ottawa: search_datasets(query="ottawa", portal="ottawa") — assert total_count > 0
6. Clean up — close the httpx.AsyncClient; CacheManager uses short-lived connections and needs no close
7. Print progress — each step prints a summary line prefixed with two spaces
8. Print "All smoke tests passed!" on success