Add interactive task examples and move call_tool_as_task to experimental

This commit adds working examples for the Tasks SEP demonstrating
elicitation and sampling flows, along with supporting infrastructure
changes.

Examples:
- simple-task-interactive server: Exposes confirm_delete (elicitation)
  and write_haiku (sampling) tools that run as tasks
- simple-task-interactive-client: Connects to server, handles callbacks,
  and demonstrates the correct task result retrieval pattern

Key changes:
- Move call_tool_as_task() from ClientSession to
  session.experimental.call_tool_as_task() for API consistency
- Add comprehensive tests mirroring the example patterns
- Add server-side print outputs for visibility into task execution

The critical insight: clients must call get_task_result() to receive
elicitation/sampling requests - simply polling get_task() will not
trigger the callbacks.

Author: maxisbey

examples/clients/simple-task-interactive-client/README.md

@@ -0,0 +1,87 @@
+# Simple Interactive Task Client
+
+A minimal MCP client demonstrating responses to interactive tasks (elicitation and sampling).
+
+## Running
+
+First, start the interactive task server in another terminal:
+
+```bash
+cd examples/servers/simple-task-interactive
+uv run mcp-simple-task-interactive
+```
+
+Then run the client:
+
+```bash
+cd examples/clients/simple-task-interactive-client
+uv run mcp-simple-task-interactive-client
+```
+
+Use `--url` to connect to a different server.
+
+## What it does
+
+1. Connects to the server via streamable HTTP
+2. Calls `confirm_delete` - server asks for confirmation, client responds via terminal
+3. Calls `write_haiku` - server requests LLM completion, client returns a hardcoded haiku
+
+## Key concepts
+
+### Elicitation callback
+
+```python
+async def elicitation_callback(context, params) -> ElicitResult:
+    # Handle user input request from server
+    return ElicitResult(action="accept", content={"confirm": True})
+```
+
+### Sampling callback
+
+```python
+async def sampling_callback(context, params) -> CreateMessageResult:
+    # Handle LLM completion request from server
+    return CreateMessageResult(model="...", role="assistant", content=...)
+```
+
+### Using call_tool_as_task
+
+```python
+# Call a tool as a task (returns immediately with task reference)
+result = await session.experimental.call_tool_as_task("tool_name", {"arg": "value"})
+task_id = result.task.taskId
+
+# Get result - this delivers elicitation/sampling requests and blocks until complete
+final = await session.experimental.get_task_result(task_id, CallToolResult)
+```
+
+**Important**: The `get_task_result()` call is what triggers the delivery of elicitation
+and sampling requests to your callbacks. It blocks until the task completes and returns
+the final result.
+
+## Expected output
+
+```text
+Available tools: ['confirm_delete', 'write_haiku']
+
+--- Demo 1: Elicitation ---
+Calling confirm_delete tool...
+Task created: <task-id>
+
+[Elicitation] Server asks: Are you sure you want to delete 'important.txt'?
+Your response (y/n): y
+[Elicitation] Responding with: confirm=True
+Result: Deleted 'important.txt'
+
+--- Demo 2: Sampling ---
+Calling write_haiku tool...
+Task created: <task-id>
+
+[Sampling] Server requests LLM completion for: Write a haiku about autumn leaves
+[Sampling] Responding with haiku
+Result:
+Haiku:
+Cherry blossoms fall
+Softly on the quiet pond
+Spring whispers goodbye
+```

examples/clients/simple-task-interactive-client/mcp_simple_task_interactive_client/__init__.py

@@ -0,0 +1,5 @@
+import sys
+
+from .main import main
+
+sys.exit(main())  # type: ignore[call-arg]

examples/clients/simple-task-interactive-client/mcp_simple_task_interactive_client/__main__.py

@@ -0,0 +1,116 @@
+"""Simple interactive task client demonstrating elicitation and sampling responses."""
+
+import asyncio
+from typing import Any
+
+import click
+from mcp import ClientSession
+from mcp.client.streamable_http import streamablehttp_client
+from mcp.shared.context import RequestContext
+from mcp.types import (
+    CallToolResult,
+    CreateMessageRequestParams,
+    CreateMessageResult,
+    ElicitRequestParams,
+    ElicitResult,
+    TextContent,
+)
+
+
+async def elicitation_callback(
+    context: RequestContext[ClientSession, Any],
+    params: ElicitRequestParams,
+) -> ElicitResult:
+    """Handle elicitation requests from the server."""
+    print(f"\n[Elicitation] Server asks: {params.message}")
+
+    # Simple terminal prompt
+    response = input("Your response (y/n): ").strip().lower()
+    confirmed = response in ("y", "yes", "true", "1")
+
+    print(f"[Elicitation] Responding with: confirm={confirmed}")
+    return ElicitResult(action="accept", content={"confirm": confirmed})
+
+
+async def sampling_callback(
+    context: RequestContext[ClientSession, Any],
+    params: CreateMessageRequestParams,
+) -> CreateMessageResult:
+    """Handle sampling requests from the server."""
+    # Get the prompt from the first message
+    prompt = "unknown"
+    if params.messages:
+        content = params.messages[0].content
+        if isinstance(content, TextContent):
+            prompt = content.text
+
+    print(f"\n[Sampling] Server requests LLM completion for: {prompt}")
+
+    # Return a hardcoded haiku (in real use, call your LLM here)
+    haiku = """Cherry blossoms fall
+Softly on the quiet pond
+Spring whispers goodbye"""
+
+    print("[Sampling] Responding with haiku")
+    return CreateMessageResult(
+        model="mock-haiku-model",
+        role="assistant",
+        content=TextContent(type="text", text=haiku),
+    )
+
+
+def get_text(result: CallToolResult) -> str:
+    """Extract text from a CallToolResult."""
+    if result.content and isinstance(result.content[0], TextContent):
+        return result.content[0].text
+    return "(no text)"
+
+
+async def run(url: str) -> None:
+    async with streamablehttp_client(url) as (read, write, _):
+        async with ClientSession(
+            read,
+            write,
+            elicitation_callback=elicitation_callback,
+            sampling_callback=sampling_callback,
+        ) as session:
+            await session.initialize()
+
+            # List tools
+            tools = await session.list_tools()
+            print(f"Available tools: {[t.name for t in tools.tools]}")
+
+            # Demo 1: Elicitation (confirm_delete)
+            print("\n--- Demo 1: Elicitation ---")
+            print("Calling confirm_delete tool...")
+
+            result = await session.experimental.call_tool_as_task("confirm_delete", {"filename": "important.txt"})
+            task_id = result.task.taskId
+            print(f"Task created: {task_id}")
+
+            # get_task_result() delivers elicitation requests and blocks until complete
+            final = await session.experimental.get_task_result(task_id, CallToolResult)
+            print(f"Result: {get_text(final)}")
+
+            # Demo 2: Sampling (write_haiku)
+            print("\n--- Demo 2: Sampling ---")
+            print("Calling write_haiku tool...")
+
+            result = await session.experimental.call_tool_as_task("write_haiku", {"topic": "autumn leaves"})
+            task_id = result.task.taskId
+            print(f"Task created: {task_id}")
+
+            # get_task_result() delivers sampling requests and blocks until complete
+            final = await session.experimental.get_task_result(task_id, CallToolResult)
+            print(f"Result:\n{get_text(final)}")
+
+
+@click.command()
+@click.option("--url", default="http://localhost:8000/mcp", help="Server URL")
+def main(url: str) -> int:
+    asyncio.run(run(url))
+    return 0
+
+
+if __name__ == "__main__":
+    main()

examples/clients/simple-task-interactive-client/mcp_simple_task_interactive_client/main.py

@@ -0,0 +1,43 @@
+[project]
+name = "mcp-simple-task-interactive-client"
+version = "0.1.0"
+description = "A simple MCP client demonstrating interactive task responses"
+readme = "README.md"
+requires-python = ">=3.10"
+authors = [{ name = "Anthropic, PBC." }]
+keywords = ["mcp", "llm", "tasks", "client", "elicitation", "sampling"]
+license = { text = "MIT" }
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+]
+dependencies = ["click>=8.0", "mcp"]
+
+[project.scripts]
+mcp-simple-task-interactive-client = "mcp_simple_task_interactive_client.main:main"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["mcp_simple_task_interactive_client"]
+
+[tool.pyright]
+include = ["mcp_simple_task_interactive_client"]
+venvPath = "."
+venv = ".venv"
+
+[tool.ruff.lint]
+select = ["E", "F", "I"]
+ignore = []
+
+[tool.ruff]
+line-length = 120
+target-version = "py310"
+
+[dependency-groups]
+dev = ["pyright>=1.1.378", "ruff>=0.6.9"]

examples/clients/simple-task-interactive-client/pyproject.toml

@@ -0,0 +1,74 @@
+# Simple Interactive Task Server
+
+A minimal MCP server demonstrating interactive tasks with elicitation and sampling.
+
+## Running
+
+```bash
+cd examples/servers/simple-task-interactive
+uv run mcp-simple-task-interactive
+```
+
+The server starts on `http://localhost:8000/mcp` by default. Use `--port` to change.
+
+## What it does
+
+This server exposes two tools:
+
+### `confirm_delete` (demonstrates elicitation)
+
+Asks the user for confirmation before "deleting" a file.
+
+- Uses `TaskSession.elicit()` to request user input
+- Shows the elicitation flow: task -> input_required -> response -> complete
+
+### `write_haiku` (demonstrates sampling)
+
+Asks the LLM to write a haiku about a topic.
+
+- Uses `TaskSession.create_message()` to request LLM completion
+- Shows the sampling flow: task -> input_required -> response -> complete
+
+## Usage with the client
+
+In one terminal, start the server:
+
+```bash
+cd examples/servers/simple-task-interactive
+uv run mcp-simple-task-interactive
+```
+
+In another terminal, run the interactive client:
+
+```bash
+cd examples/clients/simple-task-interactive-client
+uv run mcp-simple-task-interactive-client
+```
+
+## Expected server output
+
+When a client connects and calls the tools, you'll see:
+
+```text
+Starting server on http://localhost:8000/mcp
+
+[Server] confirm_delete called for 'important.txt'
+[Server] Task created: <task-id>
+[Server] Sending elicitation request to client...
+[Server] Received elicitation response: action=accept, content={'confirm': True}
+[Server] Completing task with result: Deleted 'important.txt'
+
+[Server] write_haiku called for topic 'autumn leaves'
+[Server] Task created: <task-id>
+[Server] Sending sampling request to client...
+[Server] Received sampling response: Cherry blossoms fall
+Softly on the quiet pon...
+[Server] Completing task with haiku
+```
+
+## Key concepts
+
+1. **TaskSession**: Wraps ServerSession to enqueue elicitation/sampling requests
+2. **TaskResultHandler**: Delivers queued messages and routes responses
+3. **task_execution()**: Context manager for safe task execution with auto-fail
+4. **Response routing**: Responses are routed back to waiting resolvers

examples/servers/simple-task-interactive/README.md

@@ -0,0 +1,5 @@
+import sys
+
+from .server import main
+
+sys.exit(main())  # type: ignore[call-arg]