Building MCP Servers
Core Concepts
Server
The FastMCP server is your core interface to the MCP protocol. It handles connection management, protocol compliance, and message routing:
"""Example showing lifespan support for startup/shutdown with strong typing.""" from collections.abc import AsyncIterator from contextlib import asynccontextmanager from dataclasses import dataclass from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession # Mock database class for example class Database: """Mock database class for example.""" @classmethod async def connect(cls) -> "Database": """Connect to database.""" return cls() async def disconnect(self) -> None: """Disconnect from database.""" pass def query(self) -> str: """Execute a query.""" return "Query result" @dataclass class AppContext: """Application context with typed dependencies.""" db: Database @asynccontextmanager async def app_lifespan(server: FastMCP) -> AsyncIterator[AppContext]: """Manage application lifecycle with type-safe context.""" # Initialize on startup db = await Database.connect() try: yield AppContext(db=db) finally: # Cleanup on shutdown await db.disconnect() # Pass lifespan to server mcp = FastMCP("My App", lifespan=app_lifespan) # Access type-safe lifespan context in tools @mcp.tool() def query_db(ctx: Context[ServerSession, AppContext]) -> str: """Tool that uses initialized resources.""" db = ctx.request_context.lifespan_context.db return db.query()
Full example: examples/snippets/servers/lifespan_example.py
Resources
Resources are how you expose data to LLMs. They're similar to GET endpoints in a REST API - they provide data but shouldn't perform significant computation or have side effects:
from mcp.server.fastmcp import FastMCP mcp = FastMCP(name="Resource Example") @mcp.resource("file://documents/{name}") def read_document(name: str) -> str: """Read a document by name.""" # This would normally read from disk return f"Content of {name}" @mcp.resource("config://settings") def get_settings() -> str: """Get application settings.""" return """{ "theme": "dark", "language": "en", "debug": false }"""
Full example: examples/snippets/servers/basic_resource.py
Resource Templates and Template Reading
Resources with URI parameters (e.g., {name}) are registered as templates. When a client reads a templated resource, the URI parameters are extracted and passed to the function:
from mcp.server.fastmcp import FastMCP mcp = FastMCP("Template Example") @mcp.resource("users://{user_id}/profile") def get_user_profile(user_id: str) -> str: """Read a specific user's profile. The user_id is extracted from the URI.""" return f'{{"user_id": "{user_id}", "name": "User {user_id}"}}'
Full example: examples/snippets/servers/resource_templates.py
Clients read a template resource by providing a concrete URI:
# Client-side: read a template resource with a concrete URI content = await session.read_resource("users://alice/profile")
Templates with multiple parameters work the same way:
@mcp.resource("repos://{owner}/{repo}/readme") def get_readme(owner: str, repo: str) -> str: """Each URI parameter becomes a function argument.""" return f"README for {owner}/{repo}"
Binary Resources
Resources can return binary data by returning bytes instead of str. Set the mime_type to indicate the content type:
from mcp.server.fastmcp import FastMCP mcp = FastMCP("Binary Resource Example") @mcp.resource("images://logo.png", mime_type="image/png") def get_logo() -> bytes: """Return a binary image resource.""" with open("logo.png", "rb") as f: return f.read()
Full example: examples/snippets/servers/binary_resources.py
Binary content is automatically base64-encoded and returned as BlobResourceContents in the MCP response.
Resource Subscriptions
Clients can subscribe to resource updates. Use the low-level server API to handle subscription and unsubscription requests:
from mcp.server.lowlevel import Server server = Server("Subscription Example") subscriptions: dict[str, set[str]] = {} # uri -> set of session ids @server.subscribe_resource() async def handle_subscribe(uri) -> None: """Handle a client subscribing to a resource.""" subscriptions.setdefault(str(uri), set()).add("current_session") @server.unsubscribe_resource() async def handle_unsubscribe(uri) -> None: """Handle a client unsubscribing from a resource.""" if str(uri) in subscriptions: subscriptions[str(uri)].discard("current_session")
Full example: examples/snippets/servers/resource_subscriptions.py
When a subscribed resource changes, notify clients with send_resource_updated():
from pydantic import AnyUrl # After modifying resource data: await session.send_resource_updated(AnyUrl("resource://my-resource"))
Tools
Tools let LLMs take actions through your server. Unlike resources, tools are expected to perform computation and have side effects:
from mcp.server.fastmcp import FastMCP mcp = FastMCP(name="Tool Example") @mcp.tool() def sum(a: int, b: int) -> int: """Add two numbers together.""" return a + b @mcp.tool() def get_weather(city: str, unit: str = "celsius") -> str: """Get weather for a city.""" # This would normally call a weather API return f"Weather in {city}: 22degrees{unit[0].upper()}"
Full example: examples/snippets/servers/basic_tool.py
Error Handling
When a tool encounters an error, it should signal this to the client rather than returning a normal result. The MCP protocol uses the isError flag on CallToolResult to distinguish error responses from successful ones. There are three ways to handle errors:
"""Example showing how to handle and return errors from tools.""" from mcp.server.fastmcp import FastMCP from mcp.server.fastmcp.exceptions import ToolError from mcp.types import CallToolResult, TextContent mcp = FastMCP("Tool Error Handling Example") # Option 1: Raise ToolError for expected error conditions. # The error message is returned to the client with isError=True. @mcp.tool() def divide(a: float, b: float) -> float: """Divide two numbers.""" if b == 0: raise ToolError("Cannot divide by zero") return a / b # Option 2: Unhandled exceptions are automatically caught and # converted to error responses with isError=True. @mcp.tool() def read_config(path: str) -> str: """Read a configuration file.""" # If this raises FileNotFoundError, the client receives an # error response like "Error executing tool read_config: ..." with open(path) as f: return f.read() # Option 3: Return CallToolResult directly for full control # over error responses, including custom content. @mcp.tool() def validate_input(data: str) -> CallToolResult: """Validate input data.""" errors: list[str] = [] if len(data) < 3: errors.append("Input must be at least 3 characters") if not data.isascii(): errors.append("Input must be ASCII only") if errors: return CallToolResult( content=[TextContent(type="text", text="\n".join(errors))], isError=True, ) return CallToolResult( content=[TextContent(type="text", text="Validation passed")], )
Full example: examples/snippets/servers/tool_errors.py
ToolErroris the preferred approach for most cases — raise it with a descriptive message and the framework handles the rest.- Unhandled exceptions are caught automatically, so tools won't crash the server. The exception message is forwarded to the client as an error response.
CallToolResultwithisError=Truegives full control when you need to customize the error content or include multiple content items.
Tools can optionally receive a Context object by including a parameter with the Context type annotation. This context is automatically injected by the FastMCP framework and provides access to MCP capabilities:
from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession mcp = FastMCP(name="Progress Example") @mcp.tool() async def long_running_task(task_name: str, ctx: Context[ServerSession, None], steps: int = 5) -> str: """Execute a task with progress updates.""" await ctx.info(f"Starting: {task_name}") for i in range(steps): progress = (i + 1) / steps await ctx.report_progress( progress=progress, total=1.0, message=f"Step {i + 1}/{steps}", ) await ctx.debug(f"Completed step {i + 1}") return f"Task '{task_name}' completed"
Full example: examples/snippets/servers/tool_progress.py
Structured Output
Tools will return structured results by default, if their return type annotation is compatible. Otherwise, they will return unstructured results.
Structured output supports these return types:
- Pydantic models (BaseModel subclasses)
- TypedDicts
- Dataclasses and other classes with type hints
dict[str, T](where T is any JSON-serializable type)- Primitive types (str, int, float, bool, bytes, None) - wrapped in
{"result": value} - Generic types (list, tuple, Union, Optional, etc.) - wrapped in
{"result": value}
Classes without type hints cannot be serialized for structured output. Only classes with properly annotated attributes will be converted to Pydantic models for schema generation and validation.
Structured results are automatically validated against the output schema generated from the annotation. This ensures the tool returns well-typed, validated data that clients can easily process.
Note: For backward compatibility, unstructured results are also returned. Unstructured results are provided for backward compatibility with previous versions of the MCP specification, and are quirks-compatible with previous versions of FastMCP in the current version of the SDK.
Note: In cases where a tool function's return type annotation
causes the tool to be classified as structured and this is undesirable,
the classification can be suppressed by passing structured_output=False
to the @tool decorator.
Advanced: Direct CallToolResult
For full control over tool responses including the _meta field (for passing data to client applications without exposing it to the model), you can return CallToolResult directly:
"""Example showing direct CallToolResult return for advanced control.""" from typing import Annotated from pydantic import BaseModel from mcp.server.fastmcp import FastMCP from mcp.types import CallToolResult, TextContent mcp = FastMCP("CallToolResult Example") class ValidationModel(BaseModel): """Model for validating structured output.""" status: str data: dict[str, int] @mcp.tool() def advanced_tool() -> CallToolResult: """Return CallToolResult directly for full control including _meta field.""" return CallToolResult( content=[TextContent(type="text", text="Response visible to the model")], _meta={"hidden": "data for client applications only"}, ) @mcp.tool() def validated_tool() -> Annotated[CallToolResult, ValidationModel]: """Return CallToolResult with structured output validation.""" return CallToolResult( content=[TextContent(type="text", text="Validated response")], structuredContent={"status": "success", "data": {"result": 42}}, _meta={"internal": "metadata"}, ) @mcp.tool() def empty_result_tool() -> CallToolResult: """For empty results, return CallToolResult with empty content.""" return CallToolResult(content=[])
Full example: examples/snippets/servers/direct_call_tool_result.py
Important: CallToolResult must always be returned (no Optional or Union). For empty results, use CallToolResult(content=[]). For optional simple types, use str | None without CallToolResult.
"""Example showing structured output with tools.""" from typing import TypedDict from pydantic import BaseModel, Field from mcp.server.fastmcp import FastMCP mcp = FastMCP("Structured Output Example") # Using Pydantic models for rich structured data class WeatherData(BaseModel): """Weather information structure.""" temperature: float = Field(description="Temperature in Celsius") humidity: float = Field(description="Humidity percentage") condition: str wind_speed: float @mcp.tool() def get_weather(city: str) -> WeatherData: """Get weather for a city - returns structured data.""" # Simulated weather data return WeatherData( temperature=22.5, humidity=45.0, condition="sunny", wind_speed=5.2, ) # Using TypedDict for simpler structures class LocationInfo(TypedDict): latitude: float longitude: float name: str @mcp.tool() def get_location(address: str) -> LocationInfo: """Get location coordinates""" return LocationInfo(latitude=51.5074, longitude=-0.1278, name="London, UK") # Using dict[str, Any] for flexible schemas @mcp.tool() def get_statistics(data_type: str) -> dict[str, float]: """Get various statistics""" return {"mean": 42.5, "median": 40.0, "std_dev": 5.2} # Ordinary classes with type hints work for structured output class UserProfile: name: str age: int email: str | None = None def __init__(self, name: str, age: int, email: str | None = None): self.name = name self.age = age self.email = email @mcp.tool() def get_user(user_id: str) -> UserProfile: """Get user profile - returns structured data""" return UserProfile(name="Alice", age=30, email="alice@example.com") # Classes WITHOUT type hints cannot be used for structured output class UntypedConfig: def __init__(self, setting1, setting2): # type: ignore[reportMissingParameterType] self.setting1 = setting1 self.setting2 = setting2 @mcp.tool() def get_config() -> UntypedConfig: """This returns unstructured output - no schema generated""" return UntypedConfig("value1", "value2") # Lists and other types are wrapped automatically @mcp.tool() def list_cities() -> list[str]: """Get a list of cities""" return ["London", "Paris", "Tokyo"] # Returns: {"result": ["London", "Paris", "Tokyo"]} @mcp.tool() def get_temperature(city: str) -> float: """Get temperature as a simple float""" return 22.5 # Returns: {"result": 22.5}
Full example: examples/snippets/servers/structured_output.py
Prompts
Prompts are reusable templates that help LLMs interact with your server effectively:
from mcp.server.fastmcp import FastMCP from mcp.server.fastmcp.prompts import base mcp = FastMCP(name="Prompt Example") @mcp.prompt(title="Code Review") def review_code(code: str) -> str: return f"Please review this code:\n\n{code}" @mcp.prompt(title="Debug Assistant") def debug_error(error: str) -> list[base.Message]: return [ base.UserMessage("I'm seeing this error:"), base.UserMessage(error), base.AssistantMessage("I'll help debug that. What have you tried so far?"), ]
Full example: examples/snippets/servers/basic_prompt.py
Prompts with Embedded Resources
Prompts can include embedded resources to provide file contents or data alongside the conversation messages:
import mcp.types as types from mcp.server.fastmcp import FastMCP from mcp.server.fastmcp.prompts import base mcp = FastMCP("Embedded Resource Prompt Example") @mcp.prompt() def review_file(filename: str) -> list[base.Message]: """Review a file with its contents embedded.""" file_content = open(filename).read() return [ base.UserMessage( content=types.TextContent(type="text", text=f"Please review {filename}:"), ), base.UserMessage( content=types.EmbeddedResource( type="resource", resource=types.TextResourceContents( uri=f"file://{filename}", text=file_content, mimeType="text/plain", ), ), ), ]
Full example: examples/snippets/servers/prompt_embedded_resources.py
Prompts with Image Content
Prompts can include images using ImageContent or the Image helper class:
import mcp.types as types from mcp.server.fastmcp import FastMCP from mcp.server.fastmcp.prompts import base from mcp.server.fastmcp.utilities.types import Image mcp = FastMCP("Image Prompt Example") @mcp.prompt() def describe_image(image_path: str) -> list[base.Message]: """Prompt that includes an image for analysis.""" img = Image(path=image_path) return [ base.UserMessage( content=types.TextContent(type="text", text="Describe this image:"), ), base.UserMessage( content=img.to_image_content(), ), ]
Full example: examples/snippets/servers/prompt_image_content.py
Prompt Change Notifications
When your server dynamically adds or removes prompts, notify connected clients:
from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession mcp = FastMCP("Dynamic Prompts") @mcp.tool() async def update_prompts(ctx: Context[ServerSession, None]) -> str: """Update available prompts and notify clients.""" # ... modify prompts ... await ctx.session.send_prompt_list_changed() return "Prompts updated"
Full example: examples/snippets/servers/prompt_change_notifications.py
Icons
MCP servers can provide icons for UI display. Icons can be added to the server implementation, tools, resources, and prompts:
from mcp.server.fastmcp import FastMCP, Icon # Create an icon from a file path or URL icon = Icon( src="icon.png", mimeType="image/png", sizes=["64x64"] ) # Add icons to server mcp = FastMCP( "My Server", website_url="https://example.com", icons=[icon] ) # Add icons to tools, resources, and prompts @mcp.tool(icons=[icon]) def my_tool(): """Tool with an icon.""" return "result" @mcp.resource("demo://resource", icons=[icon]) def my_resource(): """Resource with an icon.""" return "content"
Full example: examples/fastmcp/icons_demo.py
Images
FastMCP provides an Image class that automatically handles image data:
"""Example showing image handling with FastMCP.""" from PIL import Image as PILImage from mcp.server.fastmcp import FastMCP, Image mcp = FastMCP("Image Example") @mcp.tool() def create_thumbnail(image_path: str) -> Image: """Create a thumbnail from an image""" img = PILImage.open(image_path) img.thumbnail((100, 100)) return Image(data=img.tobytes(), format="png")
Full example: examples/snippets/servers/images.py
Audio
FastMCP provides an Audio class for returning audio data from tools, similar to Image:
from mcp.server.fastmcp import FastMCP from mcp.server.fastmcp.utilities.types import Audio mcp = FastMCP("Audio Example") @mcp.tool() def get_audio_from_file(file_path: str) -> Audio: """Return audio from a file path (format auto-detected from extension).""" return Audio(path=file_path) @mcp.tool() def get_audio_from_bytes(raw_audio: bytes) -> Audio: """Return audio from raw bytes with explicit format.""" return Audio(data=raw_audio, format="wav")
Full example: examples/snippets/servers/audio_example.py
The Audio class accepts path or data (mutually exclusive) and an optional format string. Supported formats include wav, mp3, ogg, flac, aac, and m4a. When using a file path, the MIME type is inferred from the file extension.
Embedded Resource Results
Tools can return EmbeddedResource to attach file contents or data inline in the result:
from mcp.server.fastmcp import FastMCP from mcp.types import EmbeddedResource, TextResourceContents mcp = FastMCP("Embedded Resource Example") @mcp.tool() def read_config(path: str) -> EmbeddedResource: """Read a config file and return it as an embedded resource.""" with open(path) as f: content = f.read() return EmbeddedResource( type="resource", resource=TextResourceContents( uri=f"file://{path}", text=content, mimeType="application/json", ), )
Full example: examples/snippets/servers/embedded_resource_results.py
For binary embedded resources, use BlobResourceContents with base64-encoded data:
import base64 from mcp.server.fastmcp import FastMCP from mcp.types import BlobResourceContents, EmbeddedResource mcp = FastMCP("Binary Embedded Resource Example") @mcp.tool() def read_binary_file(path: str) -> EmbeddedResource: """Read a binary file and return it as an embedded resource.""" with open(path, "rb") as f: data = base64.b64encode(f.read()).decode() return EmbeddedResource( type="resource", resource=BlobResourceContents( uri=f"file://{path}", blob=data, mimeType="application/octet-stream", ), )
Full example: examples/snippets/servers/embedded_resource_results_binary.py
Tool Change Notifications
When your server dynamically adds or removes tools at runtime, notify connected clients so they can refresh their tool list:
from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession mcp = FastMCP("Dynamic Tools") @mcp.tool() async def register_plugin(name: str, ctx: Context[ServerSession, None]) -> str: """Dynamically register a new tool and notify the client.""" # ... register the plugin's tools ... # Notify the client that the tool list has changed await ctx.session.send_tool_list_changed() return f"Plugin '{name}' registered"
Full example: examples/snippets/servers/tool_change_notifications.py
Context
The Context object is automatically injected into tool and resource functions that request it via type hints. It provides access to MCP capabilities like logging, progress reporting, resource reading, user interaction, and request metadata.
Getting Context in Functions
To use context in a tool or resource function, add a parameter with the Context type annotation:
from mcp.server.fastmcp import Context, FastMCP mcp = FastMCP(name="Context Example") @mcp.tool() async def my_tool(x: int, ctx: Context) -> str: """Tool that uses context capabilities.""" # The context parameter can have any name as long as it's type-annotated return await process_with_context(x, ctx)
Context Properties and Methods
The Context object provides the following capabilities:
ctx.request_id- Unique ID for the current requestctx.client_id- Client ID if availablectx.fastmcp- Access to the FastMCP server instance (see FastMCP Properties)ctx.session- Access to the underlying session for advanced communication (see Session Properties and Methods)ctx.request_context- Access to request-specific data and lifespan resources (see Request Context Properties)await ctx.debug(message)- Send debug log messageawait ctx.info(message)- Send info log messageawait ctx.warning(message)- Send warning log messageawait ctx.error(message)- Send error log messageawait ctx.log(level, message, logger_name=None)- Send log with custom levelawait ctx.report_progress(progress, total=None, message=None)- Report operation progressawait ctx.read_resource(uri)- Read a resource by URIawait ctx.elicit(message, schema)- Request additional information from user with validation
from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession mcp = FastMCP(name="Progress Example") @mcp.tool() async def long_running_task(task_name: str, ctx: Context[ServerSession, None], steps: int = 5) -> str: """Execute a task with progress updates.""" await ctx.info(f"Starting: {task_name}") for i in range(steps): progress = (i + 1) / steps await ctx.report_progress( progress=progress, total=1.0, message=f"Step {i + 1}/{steps}", ) await ctx.debug(f"Completed step {i + 1}") return f"Task '{task_name}' completed"
Full example: examples/snippets/servers/tool_progress.py
Completions
MCP supports providing completion suggestions for prompt arguments and resource template parameters. With the context parameter, servers can provide completions based on previously resolved values:
Client usage:
""" cd to the `examples/snippets` directory and run: uv run completion-client """ import asyncio import os from mcp import ClientSession, StdioServerParameters from mcp.client.stdio import stdio_client from mcp.types import PromptReference, ResourceTemplateReference # Create server parameters for stdio connection server_params = StdioServerParameters( command="uv", # Using uv to run the server args=["run", "server", "completion", "stdio"], # Server with completion support env={"UV_INDEX": os.environ.get("UV_INDEX", "")}, ) async def run(): """Run the completion client example.""" async with stdio_client(server_params) as (read, write): async with ClientSession(read, write) as session: # Initialize the connection await session.initialize() # List available resource templates templates = await session.list_resource_templates() print("Available resource templates:") for template in templates.resourceTemplates: print(f" - {template.uriTemplate}") # List available prompts prompts = await session.list_prompts() print("\nAvailable prompts:") for prompt in prompts.prompts: print(f" - {prompt.name}") # Complete resource template arguments if templates.resourceTemplates: template = templates.resourceTemplates[0] print(f"\nCompleting arguments for resource template: {template.uriTemplate}") # Complete without context result = await session.complete( ref=ResourceTemplateReference(type="ref/resource", uri=template.uriTemplate), argument={"name": "owner", "value": "model"}, ) print(f"Completions for 'owner' starting with 'model': {result.completion.values}") # Complete with context - repo suggestions based on owner result = await session.complete( ref=ResourceTemplateReference(type="ref/resource", uri=template.uriTemplate), argument={"name": "repo", "value": ""}, context_arguments={"owner": "modelcontextprotocol"}, ) print(f"Completions for 'repo' with owner='modelcontextprotocol': {result.completion.values}") # Complete prompt arguments if prompts.prompts: prompt_name = prompts.prompts[0].name print(f"\nCompleting arguments for prompt: {prompt_name}") result = await session.complete( ref=PromptReference(type="ref/prompt", name=prompt_name), argument={"name": "style", "value": ""}, ) print(f"Completions for 'style' argument: {result.completion.values}") def main(): """Entry point for the completion client.""" asyncio.run(run()) if __name__ == "__main__": main()
Full example: examples/snippets/clients/completion_client.py
Elicitation
Request additional information from users. This example shows an Elicitation during a Tool Call:
"""Elicitation examples demonstrating form and URL mode elicitation. Form mode elicitation collects structured, non-sensitive data through a schema. URL mode elicitation directs users to external URLs for sensitive operations like OAuth flows, credential collection, or payment processing. """ import uuid from pydantic import BaseModel, Field from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession from mcp.shared.exceptions import UrlElicitationRequiredError from mcp.types import ElicitRequestURLParams mcp = FastMCP(name="Elicitation Example") class BookingPreferences(BaseModel): """Schema for collecting user preferences.""" checkAlternative: bool = Field(description="Would you like to check another date?") alternativeDate: str = Field( default="2024-12-26", description="Alternative date (YYYY-MM-DD)", ) @mcp.tool() async def book_table(date: str, time: str, party_size: int, ctx: Context[ServerSession, None]) -> str: """Book a table with date availability check. This demonstrates form mode elicitation for collecting non-sensitive user input. """ # Check if date is available if date == "2024-12-25": # Date unavailable - ask user for alternative result = await ctx.elicit( message=(f"No tables available for {party_size} on {date}. Would you like to try another date?"), schema=BookingPreferences, ) if result.action == "accept" and result.data: if result.data.checkAlternative: return f"[SUCCESS] Booked for {result.data.alternativeDate}" return "[CANCELLED] No booking made" return "[CANCELLED] Booking cancelled" # Date available return f"[SUCCESS] Booked for {date} at {time}" @mcp.tool() async def secure_payment(amount: float, ctx: Context[ServerSession, None]) -> str: """Process a secure payment requiring URL confirmation. This demonstrates URL mode elicitation using ctx.elicit_url() for operations that require out-of-band user interaction. """ elicitation_id = str(uuid.uuid4()) result = await ctx.elicit_url( message=f"Please confirm payment of ${amount:.2f}", url=f"https://payments.example.com/confirm?amount={amount}&id={elicitation_id}", elicitation_id=elicitation_id, ) if result.action == "accept": # In a real app, the payment confirmation would happen out-of-band # and you'd verify the payment status from your backend return f"Payment of ${amount:.2f} initiated - check your browser to complete" elif result.action == "decline": return "Payment declined by user" return "Payment cancelled" @mcp.tool() async def connect_service(service_name: str, ctx: Context[ServerSession, None]) -> str: """Connect to a third-party service requiring OAuth authorization. This demonstrates the "throw error" pattern using UrlElicitationRequiredError. Use this pattern when the tool cannot proceed without user authorization. """ elicitation_id = str(uuid.uuid4()) # Raise UrlElicitationRequiredError to signal that the client must complete # a URL elicitation before this request can be processed. # The MCP framework will convert this to a -32042 error response. raise UrlElicitationRequiredError( [ ElicitRequestURLParams( mode="url", message=f"Authorization required to connect to {service_name}", url=f"https://{service_name}.example.com/oauth/authorize?elicit={elicitation_id}", elicitationId=elicitation_id, ) ] )
Full example: examples/snippets/servers/elicitation.py
Elicitation schemas support default values for all field types. Default values are automatically included in the JSON schema sent to clients, allowing them to pre-populate forms.
The elicit() method returns an ElicitationResult with:
action: "accept", "decline", or "cancel"data: The validated response (only when accepted)
Elicitation with Enum Values
To present a dropdown or selection list in elicitation forms, use json_schema_extra with an enum key on a str field. Do not use Literal -- use a plain str field with the enum constraint in the JSON schema:
from pydantic import BaseModel, Field from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession mcp = FastMCP("Enum Elicitation Example") class ColorPreference(BaseModel): color: str = Field( description="Pick your favorite color", json_schema_extra={"enum": ["red", "green", "blue", "yellow"]}, ) @mcp.tool() async def pick_color(ctx: Context[ServerSession, None]) -> str: """Ask the user to pick a color from a list.""" result = await ctx.elicit( message="Choose a color:", schema=ColorPreference, ) if result.action == "accept": return f"You picked: {result.data.color}" return "No color selected"
Full example: examples/snippets/servers/elicitation_enum.py
Elicitation Complete Notification
For URL mode elicitations, send a completion notification after the out-of-band interaction finishes. This tells the client that the elicitation is done and it may retry any blocked requests:
from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession mcp = FastMCP("Elicit Complete Example") @mcp.tool() async def handle_oauth_callback(elicitation_id: str, ctx: Context[ServerSession, None]) -> str: """Called when OAuth flow completes out-of-band.""" # ... process the callback ... # Notify the client that the elicitation is done await ctx.session.send_elicit_complete(elicitation_id) return "Authorization complete"
Full example: examples/snippets/servers/elicitation_complete.py
Sampling
Tools can interact with LLMs through sampling (generating text):
from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession from mcp.types import SamplingMessage, TextContent mcp = FastMCP(name="Sampling Example") @mcp.tool() async def generate_poem(topic: str, ctx: Context[ServerSession, None]) -> str: """Generate a poem using LLM sampling.""" prompt = f"Write a short poem about {topic}" result = await ctx.session.create_message( messages=[ SamplingMessage( role="user", content=TextContent(type="text", text=prompt), ) ], max_tokens=100, ) # Since we're not passing tools param, result.content is single content if result.content.type == "text": return result.content.text return str(result.content)
Full example: examples/snippets/servers/sampling.py
Logging and Notifications
Tools can send logs and notifications through the context:
from mcp.server.fastmcp import Context, FastMCP from mcp.server.session import ServerSession mcp = FastMCP(name="Notifications Example") @mcp.tool() async def process_data(data: str, ctx: Context[ServerSession, None]) -> str: """Process data with logging.""" # Different log levels await ctx.debug(f"Debug: Processing '{data}'") await ctx.info("Info: Starting processing") await ctx.warning("Warning: This is experimental") await ctx.error("Error: (This is just a demo)") # Notify about resource changes await ctx.session.send_resource_list_changed() return f"Processed: {data}"
Full example: examples/snippets/servers/notifications.py
Setting the Logging Level
Clients can request a minimum logging level via logging/setLevel. Use the low-level server API to handle this:
import mcp.types as types from mcp.server.lowlevel import Server server = Server("Logging Level Example") current_level: types.LoggingLevel = "warning" @server.set_logging_level() async def handle_set_level(level: types.LoggingLevel) -> None: """Handle client request to change the logging level.""" global current_level current_level = level
Full example: examples/snippets/servers/set_logging_level.py
When this handler is registered, the server automatically declares the logging capability during initialization.
Authentication
For OAuth 2.1 server and client authentication, see Authorization.
FastMCP Properties
The FastMCP server instance accessible via ctx.fastmcp provides access to server configuration and metadata:
ctx.fastmcp.name- The server's name as defined during initializationctx.fastmcp.instructions- Server instructions/description provided to clientsctx.fastmcp.website_url- Optional website URL for the serverctx.fastmcp.icons- Optional list of icons for UI displayctx.fastmcp.settings- Complete server configuration object containing:debug- Debug mode flaglog_level- Current logging levelhostandport- Server network configurationmount_path,sse_path,streamable_http_path- Transport pathsstateless_http- Whether the server operates in stateless mode- And other configuration options
@mcp.tool() def server_info(ctx: Context) -> dict: """Get information about the current server.""" return { "name": ctx.fastmcp.name, "instructions": ctx.fastmcp.instructions, "debug_mode": ctx.fastmcp.settings.debug, "log_level": ctx.fastmcp.settings.log_level, "host": ctx.fastmcp.settings.host, "port": ctx.fastmcp.settings.port, }
Session Properties and Methods
The session object accessible via ctx.session provides advanced control over client communication:
ctx.session.client_params- Client initialization parameters and declared capabilitiesawait ctx.session.send_log_message(level, data, logger)- Send log messages with full controlawait ctx.session.create_message(messages, max_tokens)- Request LLM sampling/completionawait ctx.session.send_progress_notification(token, progress, total, message)- Direct progress updatesawait ctx.session.send_resource_updated(uri)- Notify clients that a specific resource changedawait ctx.session.send_resource_list_changed()- Notify clients that the resource list changedawait ctx.session.send_tool_list_changed()- Notify clients that the tool list changedawait ctx.session.send_prompt_list_changed()- Notify clients that the prompt list changed
@mcp.tool() async def notify_data_update(resource_uri: str, ctx: Context) -> str: """Update data and notify clients of the change.""" # Perform data update logic here # Notify clients that this specific resource changed await ctx.session.send_resource_updated(AnyUrl(resource_uri)) # If this affects the overall resource list, notify about that too await ctx.session.send_resource_list_changed() return f"Updated {resource_uri} and notified clients"
Request Context Properties
The request context accessible via ctx.request_context contains request-specific information and resources:
ctx.request_context.lifespan_context- Access to resources initialized during server startup- Database connections, configuration objects, shared services
- Type-safe access to resources defined in your server's lifespan function
ctx.request_context.meta- Request metadata from the client including:progressToken- Token for progress notifications- Other client-provided metadata
ctx.request_context.request- The original MCP request object for advanced processingctx.request_context.request_id- Unique identifier for this request
# Example with typed lifespan context @dataclass class AppContext: db: Database config: AppConfig @mcp.tool() def query_with_config(query: str, ctx: Context) -> str: """Execute a query using shared database and configuration.""" # Access typed lifespan context app_ctx: AppContext = ctx.request_context.lifespan_context # Use shared resources connection = app_ctx.db settings = app_ctx.config # Execute query with configuration result = connection.execute(query, timeout=settings.query_timeout) return str(result)
Full lifespan example: examples/snippets/servers/lifespan_example.py
Running Your Server
Development Mode
The fastest way to test and debug your server is with the MCP Inspector:
uv run mcp dev server.py # Add dependencies uv run mcp dev server.py --with pandas --with numpy # Mount local code uv run mcp dev server.py --with-editable .
Claude Desktop Integration
Once your server is ready, install it in Claude Desktop:
uv run mcp install server.py # Custom name uv run mcp install server.py --name "My Analytics Server" # Environment variables uv run mcp install server.py -v API_KEY=abc123 -v DB_URL=postgres://... uv run mcp install server.py -f .env
Direct Execution
For advanced scenarios like custom deployments:
"""Example showing direct execution of an MCP server. This is the simplest way to run an MCP server directly. cd to the `examples/snippets` directory and run: uv run direct-execution-server or python servers/direct_execution.py """ from mcp.server.fastmcp import FastMCP mcp = FastMCP("My App") @mcp.tool() def hello(name: str = "World") -> str: """Say hello to someone.""" return f"Hello, {name}!" def main(): """Entry point for the direct execution server.""" mcp.run() if __name__ == "__main__": main()
Full example: examples/snippets/servers/direct_execution.py
Run it with:
python servers/direct_execution.py
# or
uv run mcp run servers/direct_execution.pyNote that uv run mcp run or uv run mcp dev only supports server using FastMCP and not the low-level server variant.
Streamable HTTP Transport
Note: Streamable HTTP transport is the recommended transport for production deployments. Use
stateless_http=Trueandjson_response=Truefor optimal scalability.
""" Run from the repository root: uv run examples/snippets/servers/streamable_config.py """ from mcp.server.fastmcp import FastMCP # Stateless server with JSON responses (recommended) mcp = FastMCP("StatelessServer", stateless_http=True, json_response=True) # Other configuration options: # Stateless server with SSE streaming responses # mcp = FastMCP("StatelessServer", stateless_http=True) # Stateful server with session persistence # mcp = FastMCP("StatefulServer") # Add a simple tool to demonstrate the server @mcp.tool() def greet(name: str = "World") -> str: """Greet someone by name.""" return f"Hello, {name}!" # Run server with streamable_http transport if __name__ == "__main__": mcp.run(transport="streamable-http")
Full example: examples/snippets/servers/streamable_config.py
You can mount multiple FastMCP servers in a Starlette application:
""" Run from the repository root: uvicorn examples.snippets.servers.streamable_starlette_mount:app --reload """ import contextlib from starlette.applications import Starlette from starlette.routing import Mount from mcp.server.fastmcp import FastMCP # Create the Echo server echo_mcp = FastMCP(name="EchoServer", stateless_http=True, json_response=True) @echo_mcp.tool() def echo(message: str) -> str: """A simple echo tool""" return f"Echo: {message}" # Create the Math server math_mcp = FastMCP(name="MathServer", stateless_http=True, json_response=True) @math_mcp.tool() def add_two(n: int) -> int: """Tool to add two to the input""" return n + 2 # Create a combined lifespan to manage both session managers @contextlib.asynccontextmanager async def lifespan(app: Starlette): async with contextlib.AsyncExitStack() as stack: await stack.enter_async_context(echo_mcp.session_manager.run()) await stack.enter_async_context(math_mcp.session_manager.run()) yield # Create the Starlette app and mount the MCP servers app = Starlette( routes=[ Mount("/echo", echo_mcp.streamable_http_app()), Mount("/math", math_mcp.streamable_http_app()), ], lifespan=lifespan, ) # Note: Clients connect to http://localhost:8000/echo/mcp and http://localhost:8000/math/mcp # To mount at the root of each path (e.g., /echo instead of /echo/mcp): # echo_mcp.settings.streamable_http_path = "/" # math_mcp.settings.streamable_http_path = "/"
Full example: examples/snippets/servers/streamable_starlette_mount.py
For low level server with Streamable HTTP implementations, see:
- Stateful server:
examples/servers/simple-streamablehttp/ - Stateless server:
examples/servers/simple-streamablehttp-stateless/
The streamable HTTP transport supports:
- Stateful and stateless operation modes
- Resumability with event stores
- JSON or SSE response formats
- Better scalability for multi-node deployments
CORS Configuration for Browser-Based Clients
If you'd like your server to be accessible by browser-based MCP clients, you'll need to configure CORS headers. The Mcp-Session-Id header must be exposed for browser clients to access it:
from starlette.applications import Starlette from starlette.middleware.cors import CORSMiddleware # Create your Starlette app first starlette_app = Starlette(routes=[...]) # Then wrap it with CORS middleware starlette_app = CORSMiddleware( starlette_app, allow_origins=["*"], # Configure appropriately for production allow_methods=["GET", "POST", "DELETE"], # MCP streamable HTTP methods expose_headers=["Mcp-Session-Id"], )
This configuration is necessary because:
- The MCP streamable HTTP transport uses the
Mcp-Session-Idheader for session management - Browsers restrict access to response headers unless explicitly exposed via CORS
- Without this configuration, browser-based clients won't be able to read the session ID from initialization responses
Mounting to an Existing ASGI Server
By default, SSE servers are mounted at /sse and Streamable HTTP servers are mounted at /mcp. You can customize these paths using the methods described below.
For more information on mounting applications in Starlette, see the Starlette documentation.
StreamableHTTP servers
You can mount the StreamableHTTP server to an existing ASGI server using the streamable_http_app method. This allows you to integrate the StreamableHTTP server with other ASGI applications.
Basic mounting
""" Basic example showing how to mount StreamableHTTP server in Starlette. Run from the repository root: uvicorn examples.snippets.servers.streamable_http_basic_mounting:app --reload """ import contextlib from starlette.applications import Starlette from starlette.routing import Mount from mcp.server.fastmcp import FastMCP # Create MCP server mcp = FastMCP("My App", json_response=True) @mcp.tool() def hello() -> str: """A simple hello tool""" return "Hello from MCP!" # Create a lifespan context manager to run the session manager @contextlib.asynccontextmanager async def lifespan(app: Starlette): async with mcp.session_manager.run(): yield # Mount the StreamableHTTP server to the existing ASGI server app = Starlette( routes=[ Mount("/", app=mcp.streamable_http_app()), ], lifespan=lifespan, )
Full example: examples/snippets/servers/streamable_http_basic_mounting.py
Host-based routing
""" Example showing how to mount StreamableHTTP server using Host-based routing. Run from the repository root: uvicorn examples.snippets.servers.streamable_http_host_mounting:app --reload """ import contextlib from starlette.applications import Starlette from starlette.routing import Host from mcp.server.fastmcp import FastMCP # Create MCP server mcp = FastMCP("MCP Host App", json_response=True) @mcp.tool() def domain_info() -> str: """Get domain-specific information""" return "This is served from mcp.acme.corp" # Create a lifespan context manager to run the session manager @contextlib.asynccontextmanager async def lifespan(app: Starlette): async with mcp.session_manager.run(): yield # Mount using Host-based routing app = Starlette( routes=[ Host("mcp.acme.corp", app=mcp.streamable_http_app()), ], lifespan=lifespan, )
Full example: examples/snippets/servers/streamable_http_host_mounting.py
Multiple servers with path configuration
""" Example showing how to mount multiple StreamableHTTP servers with path configuration. Run from the repository root: uvicorn examples.snippets.servers.streamable_http_multiple_servers:app --reload """ import contextlib from starlette.applications import Starlette from starlette.routing import Mount from mcp.server.fastmcp import FastMCP # Create multiple MCP servers api_mcp = FastMCP("API Server", json_response=True) chat_mcp = FastMCP("Chat Server", json_response=True) @api_mcp.tool() def api_status() -> str: """Get API status""" return "API is running" @chat_mcp.tool() def send_message(message: str) -> str: """Send a chat message""" return f"Message sent: {message}" # Configure servers to mount at the root of each path # This means endpoints will be at /api and /chat instead of /api/mcp and /chat/mcp api_mcp.settings.streamable_http_path = "/" chat_mcp.settings.streamable_http_path = "/" # Create a combined lifespan to manage both session managers @contextlib.asynccontextmanager async def lifespan(app: Starlette): async with contextlib.AsyncExitStack() as stack: await stack.enter_async_context(api_mcp.session_manager.run()) await stack.enter_async_context(chat_mcp.session_manager.run()) yield # Mount the servers app = Starlette( routes=[ Mount("/api", app=api_mcp.streamable_http_app()), Mount("/chat", app=chat_mcp.streamable_http_app()), ], lifespan=lifespan, )
Full example: examples/snippets/servers/streamable_http_multiple_servers.py
Path configuration at initialization
""" Example showing path configuration during FastMCP initialization. Run from the repository root: uvicorn examples.snippets.servers.streamable_http_path_config:app --reload """ from starlette.applications import Starlette from starlette.routing import Mount from mcp.server.fastmcp import FastMCP # Configure streamable_http_path during initialization # This server will mount at the root of wherever it's mounted mcp_at_root = FastMCP( "My Server", json_response=True, streamable_http_path="/", ) @mcp_at_root.tool() def process_data(data: str) -> str: """Process some data""" return f"Processed: {data}" # Mount at /process - endpoints will be at /process instead of /process/mcp app = Starlette( routes=[ Mount("/process", app=mcp_at_root.streamable_http_app()), ] )
Full example: examples/snippets/servers/streamable_http_path_config.py
SSE servers
Note: SSE transport is being superseded by Streamable HTTP transport.
You can mount the SSE server to an existing ASGI server using the sse_app method. This allows you to integrate the SSE server with other ASGI applications.
from starlette.applications import Starlette from starlette.routing import Mount, Host from mcp.server.fastmcp import FastMCP mcp = FastMCP("My App") # Mount the SSE server to the existing ASGI server app = Starlette( routes=[ Mount('/', app=mcp.sse_app()), ] ) # or dynamically mount as host app.router.routes.append(Host('mcp.acme.corp', app=mcp.sse_app()))
When mounting multiple MCP servers under different paths, you can configure the mount path in several ways:
from starlette.applications import Starlette from starlette.routing import Mount from mcp.server.fastmcp import FastMCP # Create multiple MCP servers github_mcp = FastMCP("GitHub API") browser_mcp = FastMCP("Browser") curl_mcp = FastMCP("Curl") search_mcp = FastMCP("Search") # Method 1: Configure mount paths via settings (recommended for persistent configuration) github_mcp.settings.mount_path = "/github" browser_mcp.settings.mount_path = "/browser" # Method 2: Pass mount path directly to sse_app (preferred for ad-hoc mounting) # This approach doesn't modify the server's settings permanently # Create Starlette app with multiple mounted servers app = Starlette( routes=[ # Using settings-based configuration Mount("/github", app=github_mcp.sse_app()), Mount("/browser", app=browser_mcp.sse_app()), # Using direct mount path parameter Mount("/curl", app=curl_mcp.sse_app("/curl")), Mount("/search", app=search_mcp.sse_app("/search")), ] ) # Method 3: For direct execution, you can also pass the mount path to run() if __name__ == "__main__": search_mcp.run(transport="sse", mount_path="/search")
For more information on mounting applications in Starlette, see the Starlette documentation.
Advanced Usage
For the low-level server API, pagination, and direct handler registration, see Low-Level Server.