Send logging output and progress notifications from a Python MCP tool (Context object)

Documented steps

1. Get a Context by annotating any tool parameter with it: from mcp.server.mcpserver import Context; def search(query: str, ctx: Context) -> str — injection is by type annotation, the parameter never appears in the tool's JSON input schema and clients can't fill it.
2. Logging: use plain Python logging (logger = logging.getLogger(__name__); logger.info(...)). MCP's protocol-level logging capability (pushing log notifications through Context) is deprecated by the 2026-07-28 spec revision without replacement — don't build on ctx-based logging. Docs: https://py.sdk.modelcontextprotocol.io/handlers/logging/
3. MCPServer("Name", log_level="DEBUG") wires logging.basicConfig() at construction (default INFO), but leaves any logging config you set up first untouched. On stdio, logging's default handler writes to stderr — safe; stdout is the wire.
4. Progress: from an async def tool call await ctx.report_progress(progress, total=None, message=None), e.g. per loop iteration: await ctx.report_progress(done, total=len(urls), message=f"Imported {url}"). Omit total when the denominator is unknown. Docs: https://py.sdk.modelcontextprotocol.io/handlers/progress/
5. Client side: progress is opted into per call, not per client: await client.call_tool("import_catalog", {...}, progress_callback=show) where show is async def show(progress, total, message). Each report_progress call becomes one callback invocation, streamed while the tool runs.