"""FastMCP - A more ergonomic interface for MCP servers."""
from __future__ import annotations
import asyncio
import inspect
import re
import secrets
import warnings
import weakref
from collections.abc import (
AsyncIterator,
Awaitable,
Callable,
Collection,
Mapping,
Sequence,
)
from contextlib import (
AbstractAsyncContextManager,
AsyncExitStack,
asynccontextmanager,
suppress,
)
from dataclasses import dataclass
from functools import partial
from pathlib import Path
from typing import TYPE_CHECKING, Any, Generic, Literal, cast, overload
import anyio
import httpx
import mcp.types
import uvicorn
from docket import Docket, Worker
from mcp.server.lowlevel.helper_types import ReadResourceContents
from mcp.server.lowlevel.server import LifespanResultT, NotificationOptions
from mcp.server.stdio import stdio_server
from mcp.shared.exceptions import McpError
from mcp.types import (
METHOD_NOT_FOUND,
Annotations,
AnyFunction,
CallToolRequestParams,
ContentBlock,
ErrorData,
GetPromptResult,
ToolAnnotations,
)
from mcp.types import Prompt as SDKPrompt
from mcp.types import Resource as SDKResource
from mcp.types import ResourceTemplate as SDKResourceTemplate
from mcp.types import Tool as SDKTool
from pydantic import AnyUrl
from starlette.middleware import Middleware as ASGIMiddleware
from starlette.requests import Request
from starlette.responses import Response
from starlette.routing import BaseRoute, Route
import fastmcp
import fastmcp.server
from fastmcp.exceptions import DisabledError, NotFoundError
from fastmcp.mcp_config import MCPConfig
from fastmcp.prompts import Prompt
from fastmcp.prompts.prompt import FunctionPrompt
from fastmcp.prompts.prompt_manager import PromptManager
from fastmcp.resources.resource import FunctionResource, Resource
from fastmcp.resources.resource_manager import ResourceManager
from fastmcp.resources.template import FunctionResourceTemplate, ResourceTemplate
from fastmcp.server.auth import AuthProvider
from fastmcp.server.event_store import EventStore
from fastmcp.server.http import (
StarletteWithLifespan,
create_sse_app,
create_streamable_http_app,
)
from fastmcp.server.low_level import LowLevelServer
from fastmcp.server.middleware import Middleware, MiddlewareContext
from fastmcp.server.tasks.capabilities import get_task_capabilities
from fastmcp.server.tasks.config import TaskConfig
from fastmcp.server.tasks.handlers import (
handle_prompt_as_task,
handle_resource_as_task,
handle_tool_as_task,
)
from fastmcp.settings import Settings
from fastmcp.tools.tool import FunctionTool, Tool, ToolResult
from fastmcp.tools.tool_manager import ToolManager
from fastmcp.tools.tool_transform import ToolTransformConfig
from fastmcp.utilities.cli import log_server_banner
from fastmcp.utilities.components import FastMCPComponent
from fastmcp.utilities.logging import get_logger, temporary_log_level
from fastmcp.utilities.types import NotSet, NotSetT
if TYPE_CHECKING:
from fastmcp.client import Client
from fastmcp.client.client import FastMCP1Server
from fastmcp.client.sampling import SamplingHandler
from fastmcp.client.transports import ClientTransport, ClientTransportT
from fastmcp.server.openapi import ComponentFn as OpenAPIComponentFn
from fastmcp.server.openapi import FastMCPOpenAPI, RouteMap
from fastmcp.server.openapi import RouteMapFn as OpenAPIRouteMapFn
from fastmcp.server.proxy import FastMCPProxy
from fastmcp.tools.tool import ToolResultSerializerType
logger = get_logger(__name__)
DuplicateBehavior = Literal["warn", "error", "replace", "ignore"]
Transport = Literal["stdio", "http", "sse", "streamable-http"]
# Compiled URI parsing regex to split a URI into protocol and path components
URI_PATTERN = re.compile(r"^([^:]+://)(.*?)$")
LifespanCallable = Callable[
["FastMCP[LifespanResultT]"], AbstractAsyncContextManager[LifespanResultT]
]
@asynccontextmanager
async def default_lifespan(server: FastMCP[LifespanResultT]) -> AsyncIterator[Any]:
"""Default lifespan context manager that does nothing.
Args:
server: The server instance this lifespan is managing
Returns:
An empty dictionary as the lifespan result.
"""
yield {}
def _lifespan_proxy(
fastmcp_server: FastMCP[LifespanResultT],
) -> Callable[
[LowLevelServer[LifespanResultT]], AbstractAsyncContextManager[LifespanResultT]
]:
@asynccontextmanager
async def wrap(
low_level_server: LowLevelServer[LifespanResultT],
) -> AsyncIterator[LifespanResultT]:
if fastmcp_server._lifespan is default_lifespan:
yield {}
return
if not fastmcp_server._lifespan_result_set:
raise RuntimeError(
"FastMCP server has a lifespan defined but no lifespan result is set, which means the server's context manager was not entered. "
+ " Are you running the server in a way that supports lifespans? If so, please file an issue at https://github.com/jlowin/fastmcp/issues."
)
yield fastmcp_server._lifespan_result
return wrap
class FastMCP(Generic[LifespanResultT]):
def __init__(
self,
name: str | None = None,
instructions: str | None = None,
*,
version: str | None = None,
website_url: str | None = None,
icons: list[mcp.types.Icon] | None = None,
auth: AuthProvider | NotSetT | None = NotSet,
middleware: Sequence[Middleware] | None = None,
lifespan: LifespanCallable | None = None,
mask_error_details: bool | None = None,
tools: Sequence[Tool | Callable[..., Any]] | None = None,
tool_transformations: Mapping[str, ToolTransformConfig] | None = None,
tool_serializer: ToolResultSerializerType | None = None,
include_tags: Collection[str] | None = None,
exclude_tags: Collection[str] | None = None,
include_fastmcp_meta: bool | None = None,
on_duplicate_tools: DuplicateBehavior | None = None,
on_duplicate_resources: DuplicateBehavior | None = None,
on_duplicate_prompts: DuplicateBehavior | None = None,
strict_input_validation: bool | None = None,
tasks: bool | None = None,
# ---
# ---
# --- The following arguments are DEPRECATED ---
# ---
# ---
log_level: str | None = None,
debug: bool | None = None,
host: str | None = None,
port: int | None = None,
sse_path: str | None = None,
message_path: str | None = None,
streamable_http_path: str | None = None,
json_response: bool | None = None,
stateless_http: bool | None = None,
sampling_handler: SamplingHandler | None = None,
sampling_handler_behavior: Literal["always", "fallback"] | None = None,
):
# Resolve server default for background task support
self._support_tasks_by_default: bool = tasks if tasks is not None else False
# Docket instance (set during lifespan for cross-task access)
self._docket = None
self._additional_http_routes: list[BaseRoute] = []
self._mounted_servers: list[MountedServer] = []
self._is_mounted: bool = (
False # Set to True when this server is mounted on another
)
self._tool_manager: ToolManager = ToolManager(
duplicate_behavior=on_duplicate_tools,
mask_error_details=mask_error_details,
transformations=tool_transformations,
)
self._resource_manager: ResourceManager = ResourceManager(
duplicate_behavior=on_duplicate_resources,
mask_error_details=mask_error_details,
)
self._prompt_manager: PromptManager = PromptManager(
duplicate_behavior=on_duplicate_prompts,
mask_error_details=mask_error_details,
)
self._tool_serializer: Callable[[Any], str] | None = tool_serializer
self._lifespan: LifespanCallable[LifespanResultT] = lifespan or default_lifespan
self._lifespan_result: LifespanResultT | None = None
self._lifespan_result_set: bool = False
self._started: asyncio.Event = asyncio.Event()
# Generate random ID if no name provided
self._mcp_server: LowLevelServer[LifespanResultT, Any] = LowLevelServer[
LifespanResultT
](
fastmcp=self,
name=name or self.generate_name(),
version=version or fastmcp.__version__,
instructions=instructions,
website_url=website_url,
icons=icons,
lifespan=_lifespan_proxy(fastmcp_server=self),
)
# if auth is `NotSet`, try to create a provider from the environment
if auth is NotSet:
if fastmcp.settings.server_auth is not None:
# server_auth_class returns the class itself
auth = fastmcp.settings.server_auth_class()
else:
auth = None
self.auth: AuthProvider | None = cast(AuthProvider | None, auth)
if tools:
for tool in tools:
if not isinstance(tool, Tool):
tool = Tool.from_function(tool, serializer=self._tool_serializer)
self.add_tool(tool)
self.include_tags: set[str] | None = (
set(include_tags) if include_tags is not None else None
)
self.exclude_tags: set[str] | None = (
set(exclude_tags) if exclude_tags is not None else None
)
self.strict_input_validation: bool = (
strict_input_validation
if strict_input_validation is not None
else fastmcp.settings.strict_input_validation
)
self.middleware: list[Middleware] = list(middleware or [])
# Set up MCP protocol handlers
self._setup_handlers()
self.sampling_handler: SamplingHandler | None = sampling_handler
self.sampling_handler_behavior: Literal["always", "fallback"] = (
sampling_handler_behavior or "fallback"
)
self.include_fastmcp_meta: bool = (
include_fastmcp_meta
if include_fastmcp_meta is not None
else fastmcp.settings.include_fastmcp_meta
)
self._handle_deprecated_settings(
log_level=log_level,
debug=debug,
host=host,
port=port,
sse_path=sse_path,
message_path=message_path,
streamable_http_path=streamable_http_path,
json_response=json_response,
stateless_http=stateless_http,
)
def __repr__(self) -> str:
return f"{type(self).__name__}({self.name!r})"
def _handle_deprecated_settings(
self,
log_level: str | None,
debug: bool | None,
host: str | None,
port: int | None,
sse_path: str | None,
message_path: str | None,
streamable_http_path: str | None,
json_response: bool | None,
stateless_http: bool | None,
) -> None:
"""Handle deprecated settings. Deprecated in 2.8.0."""
deprecated_settings: dict[str, Any] = {}
for name, arg in [
("log_level", log_level),
("debug", debug),
("host", host),
("port", port),
("sse_path", sse_path),
("message_path", message_path),
("streamable_http_path", streamable_http_path),
("json_response", json_response),
("stateless_http", stateless_http),
]:
if arg is not None:
# Deprecated in 2.8.0
if fastmcp.settings.deprecation_warnings:
warnings.warn(
f"Providing `{name}` when creating a server is deprecated. Provide it when calling `run` or as a global setting instead.",
DeprecationWarning,
stacklevel=2,
)
deprecated_settings[name] = arg
combined_settings = fastmcp.settings.model_dump() | deprecated_settings
self._deprecated_settings = Settings(**combined_settings)
@property
def settings(self) -> Settings:
# Deprecated in 2.8.0
if fastmcp.settings.deprecation_warnings:
warnings.warn(
"Accessing `.settings` on a FastMCP instance is deprecated. Use the global `fastmcp.settings` instead.",
DeprecationWarning,
stacklevel=2,
)
return self._deprecated_settings
@property
def name(self) -> str:
return self._mcp_server.name
@property
def instructions(self) -> str | None:
return self._mcp_server.instructions
@instructions.setter
def instructions(self, value: str | None) -> None:
self._mcp_server.instructions = value
@property
def version(self) -> str | None:
return self._mcp_server.version
@property
def website_url(self) -> str | None:
return self._mcp_server.website_url
@property
def icons(self) -> list[mcp.types.Icon]:
if self._mcp_server.icons is None:
return []
else:
return list(self._mcp_server.icons)
@property
def docket(self) -> Docket | None:
"""Get the Docket instance if Docket support is enabled.
Returns None if Docket is not enabled or server hasn't been started yet.
"""
return self._docket
@asynccontextmanager
async def _docket_lifespan(self) -> AsyncIterator[None]:
"""Manage Docket instance and Worker for background task execution."""
from fastmcp import settings
# Set FastMCP server in ContextVar so CurrentFastMCP can access it (use weakref to avoid reference cycles)
from fastmcp.server.dependencies import (
_current_docket,
_current_server,
_current_worker,
)
server_token = _current_server.set(weakref.ref(self))
try:
# For directly mounted servers, the parent's Docket/Worker handles all
# task execution. Skip creating our own to avoid race conditions with
# multiple workers competing for tasks from the same queue.
if self._is_mounted:
yield
return
# Create Docket instance using configured name and URL
async with Docket(
name=settings.docket.name,
url=settings.docket.url,
) as docket:
# Store on server instance for cross-task access (FastMCPTransport)
self._docket = docket
# Register local task-enabled tools/prompts/resources with Docket
# Only function-based variants support background tasks
# Register components where task execution is not "forbidden"
for tool in self._tool_manager._tools.values():
if (
isinstance(tool, FunctionTool)
and tool.task_config.mode != "forbidden"
):
docket.register(tool.fn, names=[tool.key])
for prompt in self._prompt_manager._prompts.values():
if (
isinstance(prompt, FunctionPrompt)
and prompt.task_config.mode != "forbidden"
):
# task execution requires async fn (validated at creation time)
docket.register(
cast(Callable[..., Awaitable[Any]], prompt.fn),
names=[prompt.key],
)
for resource in self._resource_manager._resources.values():
if (
isinstance(resource, FunctionResource)
and resource.task_config.mode != "forbidden"
):
docket.register(resource.fn, names=[resource.name])
for template in self._resource_manager._templates.values():
if (
isinstance(template, FunctionResourceTemplate)
and template.task_config.mode != "forbidden"
):
docket.register(template.fn, names=[template.name])
# Also register functions from mounted servers so tasks can
# execute in the parent's Docket context
for mounted in self._mounted_servers:
await self._register_mounted_server_functions(
mounted.server, docket, mounted.prefix, mounted.tool_names
)
# Set Docket in ContextVar so CurrentDocket can access it
docket_token = _current_docket.set(docket)
try:
# Build worker kwargs from settings
worker_kwargs: dict[str, Any] = {
"concurrency": settings.docket.concurrency,
"redelivery_timeout": settings.docket.redelivery_timeout,
"reconnection_delay": settings.docket.reconnection_delay,
}
if settings.docket.worker_name:
worker_kwargs["name"] = settings.docket.worker_name
# Create and start Worker
async with Worker(docket, **worker_kwargs) as worker: # type: ignore[arg-type]
# Set Worker in ContextVar so CurrentWorker can access it
worker_token = _current_worker.set(worker)
try:
worker_task = asyncio.create_task(worker.run_forever())
try:
yield
finally:
worker_task.cancel()
with suppress(asyncio.CancelledError):
await worker_task
finally:
_current_worker.reset(worker_token)
finally:
# Reset ContextVar
_current_docket.reset(docket_token)
# Clear instance attribute
self._docket = None
finally:
# Reset server ContextVar
_current_server.reset(server_token)
async def _register_mounted_server_functions(
self,
server: FastMCP,
docket: Docket,
prefix: str | None,
tool_names: dict[str, str] | None = None,
) -> None:
"""Register task-enabled functions from a mounted server with Docket.
This enables background task execution for mounted server components
through the parent server's Docket context.
Args:
server: The mounted server whose functions to register
docket: The Docket instance to register with
prefix: The mount prefix to prepend to function names (matches
client-facing tool/prompt names)
tool_names: Optional mapping of original tool names to custom names
"""
# Register tools with prefixed names to avoid collisions
for tool in server._tool_manager._tools.values():
if isinstance(tool, FunctionTool) and tool.task_config.mode != "forbidden":
# Apply tool_names override first, then prefix (matches get_tools logic)
if tool_names and tool.key in tool_names:
fn_name = tool_names[tool.key]
elif prefix:
fn_name = f"{prefix}_{tool.key}"
else:
fn_name = tool.key
docket.register(tool.fn, names=[fn_name])
# Register prompts with prefixed names
for prompt in server._prompt_manager._prompts.values():
if (
isinstance(prompt, FunctionPrompt)
and prompt.task_config.mode != "forbidden"
):
fn_name = f"{prefix}_{prompt.key}" if prefix else prompt.key
docket.register(
cast(Callable[..., Awaitable[Any]], prompt.fn),
names=[fn_name],
)
# Register resources with prefixed names (use name, not key/URI)
for resource in server._resource_manager._resources.values():
if (
isinstance(resource, FunctionResource)
and resource.task_config.mode != "forbidden"
):
fn_name = f"{prefix}_{resource.name}" if prefix else resource.name
docket.register(resource.fn, names=[fn_name])
# Register resource templates with prefixed names (use name, not key/URI)
for template in server._resource_manager._templates.values():
if (
isinstance(template, FunctionResourceTemplate)
and template.task_config.mode != "forbidden"
):
fn_name = f"{prefix}_{template.name}" if prefix else template.name
docket.register(template.fn, names=[fn_name])
# Recursively register from nested mounted servers with accumulated prefix
for nested in server._mounted_servers:
nested_prefix = (
f"{prefix}_{nested.prefix}"
if prefix and nested.prefix
else (prefix or nested.prefix)
)
await self._register_mounted_server_functions(
nested.server, docket, nested_prefix, nested.tool_names
)
@asynccontextmanager
async def _lifespan_manager(self) -> AsyncIterator[None]:
if self._lifespan_result_set:
yield
return
async with (
self._lifespan(self) as user_lifespan_result,
self._docket_lifespan(),
):
self._lifespan_result = user_lifespan_result
self._lifespan_result_set = True
async with AsyncExitStack[bool | None]() as stack:
for server in self._mounted_servers:
await stack.enter_async_context(
cm=server.server._lifespan_manager()
)
self._started.set()
try:
yield
finally:
self._started.clear()
self._lifespan_result_set = False
self._lifespan_result = None
async def run_async(
self,
transport: Transport | None = None,
show_banner: bool = True,
**transport_kwargs: Any,
) -> None:
"""Run the FastMCP server asynchronously.
Args:
transport: Transport protocol to use ("stdio", "sse", or "streamable-http")
"""
if transport is None:
transport = "stdio"
if transport not in {"stdio", "http", "sse", "streamable-http"}:
raise ValueError(f"Unknown transport: {transport}")
if transport == "stdio":
await self.run_stdio_async(
show_banner=show_banner,
**transport_kwargs,
)
elif transport in {"http", "sse", "streamable-http"}:
await self.run_http_async(
transport=transport,
show_banner=show_banner,
**transport_kwargs,
)
else:
raise ValueError(f"Unknown transport: {transport}")
def run(
self,
transport: Transport | None = None,
show_banner: bool = True,
**transport_kwargs: Any,
) -> None:
"""Run the FastMCP server. Note this is a synchronous function.
Args:
transport: Transport protocol to use ("stdio", "sse", or "streamable-http")
"""
anyio.run(
partial(
self.run_async,
transport,
show_banner=show_banner,
**transport_kwargs,
)
)
def _setup_handlers(self) -> None:
"""Set up core MCP protocol handlers."""
self._mcp_server.list_tools()(self._list_tools_mcp)
self._mcp_server.list_resources()(self._list_resources_mcp)
self._mcp_server.list_resource_templates()(self._list_resource_templates_mcp)
self._mcp_server.list_prompts()(self._list_prompts_mcp)
self._mcp_server.call_tool(validate_input=self.strict_input_validation)(
self._call_tool_mcp
)
# Register custom read_resource handler (SDK decorator doesn't support CreateTaskResult)
self._setup_read_resource_handler()
# Register custom get_prompt handler (SDK decorator doesn't support CreateTaskResult)
self._setup_get_prompt_handler()
# Register custom SEP-1686 task protocol handlers
self._setup_task_protocol_handlers()
def _setup_read_resource_handler(self) -> None:
"""
Set up custom read_resource handler that supports task-augmented responses.
The SDK's read_resource decorator doesn't support CreateTaskResult returns,
so we register a custom handler that checks request_context.experimental.is_task.
"""
async def handler(req: mcp.types.ReadResourceRequest) -> mcp.types.ServerResult:
uri = req.params.uri
# Check for task metadata via SDK's request context
task_meta = None
try:
ctx = self._mcp_server.request_context
if ctx.experimental.is_task:
task_meta = ctx.experimental.task_metadata
except (AttributeError, LookupError):
pass
# Check for task metadata and route appropriately
async with fastmcp.server.context.Context(fastmcp=self):
# Get resource including from mounted servers
resource = await self._get_resource_or_template_or_none(str(uri))
if (
resource
and self._should_enable_component(resource)
and hasattr(resource, "task_config")
):
task_mode = resource.task_config.mode # type: ignore[union-attr]
# Enforce mode="required" - must have task metadata
if task_mode == "required" and not task_meta:
raise McpError(
ErrorData(
code=METHOD_NOT_FOUND,
message=f"Resource '{uri}' requires task-augmented execution",
)
)
# Route to background if task metadata present and mode allows
if task_meta and task_mode != "forbidden":
# For FunctionResource/FunctionResourceTemplate, use Docket
if isinstance(
resource,
FunctionResource | FunctionResourceTemplate,
):
task_meta_dict = task_meta.model_dump(exclude_none=True)
return await handle_resource_as_task(
self, str(uri), resource, task_meta_dict
)
# Forbidden mode: task requested but mode="forbidden"
# Raise error since resources don't have isError field
if task_meta and task_mode == "forbidden":
raise McpError(
ErrorData(
code=METHOD_NOT_FOUND,
message=f"Resource '{uri}' does not support task-augmented execution",
)
)
# Synchronous execution
result = await self._read_resource_mcp(uri)
# Graceful degradation: if we got here with task_meta, something went wrong
# (This should be unreachable now that forbidden raises)
if task_meta:
mcp_contents = []
for item in result:
if isinstance(item.content, str):
mcp_contents.append(
mcp.types.TextResourceContents(
uri=uri,
text=item.content,
mimeType=item.mime_type or "text/plain",
)
)
elif isinstance(item.content, bytes):
import base64
mcp_contents.append(
mcp.types.BlobResourceContents(
uri=uri,
blob=base64.b64encode(item.content).decode(),
mimeType=item.mime_type or "application/octet-stream",
)
)
return mcp.types.ServerResult(
mcp.types.ReadResourceResult(
contents=mcp_contents,
_meta={
"modelcontextprotocol.io/task": {
"returned_immediately": True
}
},
)
)
# Convert to proper ServerResult
if isinstance(result, mcp.types.ServerResult):
return result
mcp_contents = []
for item in result:
if isinstance(item.content, str):
mcp_contents.append(
mcp.types.TextResourceContents(
uri=uri,
text=item.content,
mimeType=item.mime_type or "text/plain",
)
)
elif isinstance(item.content, bytes):
import base64
mcp_contents.append(
mcp.types.BlobResourceContents(
uri=uri,
blob=base64.b64encode(item.content).decode(),
mimeType=item.mime_type or "application/octet-stream",
)
)
return mcp.types.ServerResult(
mcp.types.ReadResourceResult(contents=mcp_contents)
)
self._mcp_server.request_handlers[mcp.types.ReadResourceRequest] = handler
def _setup_get_prompt_handler(self) -> None:
"""
Set up custom get_prompt handler that supports task-augmented responses.
The SDK's get_prompt decorator doesn't support CreateTaskResult returns,
so we register a custom handler that checks request_context.experimental.is_task.
"""
async def handler(req: mcp.types.GetPromptRequest) -> mcp.types.ServerResult:
name = req.params.name
arguments = req.params.arguments
# Check for task metadata via SDK's request context
task_meta = None
try:
ctx = self._mcp_server.request_context
if ctx.experimental.is_task:
task_meta = ctx.experimental.task_metadata
except (AttributeError, LookupError):
pass
# Check for task metadata and route appropriately
async with fastmcp.server.context.Context(fastmcp=self):
prompts = await self.get_prompts()
prompt = prompts.get(name)
if (
prompt
and self._should_enable_component(prompt)
and hasattr(prompt, "task_config")
and prompt.task_config
):
task_mode = prompt.task_config.mode # type: ignore[union-attr]
# Enforce mode="required" - must have task metadata
if task_mode == "required" and not task_meta:
raise McpError(
ErrorData(
code=METHOD_NOT_FOUND,
message=f"Prompt '{name}' requires task-augmented execution",
)
)
# Route to background if task metadata present and mode allows
if task_meta and task_mode != "forbidden":
task_meta_dict = task_meta.model_dump(exclude_none=True)
result = await handle_prompt_as_task(
self, name, arguments, task_meta_dict
)
return mcp.types.ServerResult(result)
# Forbidden mode: task requested but mode="forbidden"
# Raise error since prompts don't have isError field
if task_meta and task_mode == "forbidden":
raise McpError(
ErrorData(
code=METHOD_NOT_FOUND,
message=f"Prompt '{name}' does not support task-augmented execution",
)
)
# Synchronous execution
result = await self._get_prompt_mcp(name, arguments)
return mcp.types.ServerResult(result)
self._mcp_server.request_handlers[mcp.types.GetPromptRequest] = handler
def _setup_task_protocol_handlers(self) -> None:
"""Register SEP-1686 task protocol handlers with SDK."""
from mcp.types import (
CancelTaskRequest,
GetTaskPayloadRequest,
GetTaskRequest,
ListTasksRequest,
ServerResult,
)
from fastmcp.server.tasks.protocol import (
tasks_cancel_handler,
tasks_get_handler,
tasks_list_handler,
tasks_result_handler,
)
# Manually register handlers (SDK decorators fail with locally-defined functions)
# SDK expects handlers that receive Request objects and return ServerResult
async def handle_get_task(req: GetTaskRequest) -> ServerResult:
params = req.params.model_dump(by_alias=True, exclude_none=True)
result = await tasks_get_handler(self, params)
return ServerResult(result)
async def handle_get_task_result(req: GetTaskPayloadRequest) -> ServerResult:
params = req.params.model_dump(by_alias=True, exclude_none=True)
result = await tasks_result_handler(self, params)
return ServerResult(result)
async def handle_list_tasks(req: ListTasksRequest) -> ServerResult:
params = (
req.params.model_dump(by_alias=True, exclude_none=True)
if req.params
else {}
)
result = await tasks_list_handler(self, params)
return ServerResult(result)
async def handle_cancel_task(req: CancelTaskRequest) -> ServerResult:
params = req.params.model_dump(by_alias=True, exclude_none=True)
result = await tasks_cancel_handler(self, params)
return ServerResult(result)
# Register directly with SDK (same as what decorators do internally)
self._mcp_server.request_handlers[GetTaskRequest] = handle_get_task
self._mcp_server.request_handlers[GetTaskPayloadRequest] = (
handle_get_task_result
)
self._mcp_server.request_handlers[ListTasksRequest] = handle_list_tasks
self._mcp_server.request_handlers[CancelTaskRequest] = handle_cancel_task
async def _apply_middleware(
self,
context: MiddlewareContext[Any],
call_next: Callable[[MiddlewareContext[Any]], Awaitable[Any]],
) -> Any:
"""Builds and executes the middleware chain."""
chain = call_next
for mw in reversed(self.middleware):
chain = partial(mw, call_next=chain)
return await chain(context)
def add_middleware(self, middleware: Middleware) -> None:
self.middleware.append(middleware)
async def get_tools(self) -> dict[str, Tool]:
"""Get all tools (unfiltered), including mounted servers, indexed by key."""
all_tools = dict(await self._tool_manager.get_tools())
for mounted in self._mounted_servers:
try:
child_tools = await mounted.server.get_tools()
for key, tool in child_tools.items():
# Check for manual override first, then apply prefix
if mounted.tool_names and key in mounted.tool_names:
new_key = mounted.tool_names[key]
elif mounted.prefix:
new_key = f"{mounted.prefix}_{key}"
else:
new_key = key
all_tools[new_key] = tool.model_copy(key=new_key)
except Exception as e:
logger.warning(
f"Failed to get tools from mounted server {mounted.server.name!r}: {e}"
)
if fastmcp.settings.mounted_components_raise_on_load_error:
raise
continue
return all_tools
async def get_tool(self, key: str) -> Tool:
tools = await self.get_tools()
if key not in tools:
raise NotFoundError(f"Unknown tool: {key}")
return tools[key]
async def _get_tool_with_task_config(self, key: str) -> Tool | None:
"""Get a tool by key, returning None if not found.
Used for task config checking where we need the actual tool object
(including from mounted servers and proxies) but don't want to raise.
"""
try:
return await self.get_tool(key)
except NotFoundError:
return None
async def _get_resource_or_template_or_none(
self, uri: str
) -> Resource | ResourceTemplate | None:
"""Get a resource or template by URI, searching recursively. Returns None if not found."""
try:
return await self.get_resource(uri)
except NotFoundError:
pass
templates = await self.get_resource_templates()
for template in templates.values():
if template.matches(uri):
return template
return None
async def get_resources(self) -> dict[str, Resource]:
"""Get all resources (unfiltered), including mounted servers, indexed by key."""
all_resources = dict(await self._resource_manager.get_resources())
for mounted in self._mounted_servers:
try:
child_resources = await mounted.server.get_resources()
for key, resource in child_resources.items():
new_key = (
add_resource_prefix(key, mounted.prefix)
if mounted.prefix
else key
)
update = (
{"name": f"{mounted.prefix}_{resource.name}"}
if mounted.prefix and resource.name
else {}
)
all_resources[new_key] = resource.model_copy(
key=new_key, update=update
)
except Exception as e:
logger.warning(
f"Failed to get resources from mounted server {mounted.server.name!r}: {e}"
)
if fastmcp.settings.mounted_components_raise_on_load_error:
raise
continue
return all_resources
async def get_resource(self, key: str) -> Resource:
resources = await self.get_resources()
if key not in resources:
raise NotFoundError(f"Unknown resource: {key}")
return resources[key]
async def get_resource_templates(self) -> dict[str, ResourceTemplate]:
"""Get all resource templates (unfiltered), including mounted servers, indexed by key."""
all_templates = dict(await self._resource_manager.get_resource_templates())
for mounted in self._mounted_servers:
try:
child_templates = await mounted.server.get_resource_templates()
for key, template in child_templates.items():
new_key = (
add_resource_prefix(key, mounted.prefix)
if mounted.prefix
else key
)
update: dict[str, Any] = {}
if mounted.prefix:
if template.name:
update["name"] = f"{mounted.prefix}_{template.name}"
# Update uri_template so matches() works with prefixed URIs
update["uri_template"] = new_key
all_templates[new_key] = template.model_copy(
key=new_key, update=update
)
except Exception as e:
logger.warning(
f"Failed to get resource templates from mounted server {mounted.server.name!r}: {e}"
)
if fastmcp.settings.mounted_components_raise_on_load_error:
raise
continue
return all_templates
async def get_resource_template(self, key: str) -> ResourceTemplate:
"""Get a registered resource template by key."""
templates = await self.get_resource_templates()
if key not in templates:
raise NotFoundError(f"Unknown resource template: {key}")
return templates[key]
async def get_prompts(self) -> dict[str, Prompt]:
"""Get all prompts (unfiltered), including mounted servers, indexed by key."""
all_prompts = dict(await self._prompt_manager.get_prompts())
for mounted in self._mounted_servers:
try:
child_prompts = await mounted.server.get_prompts()
for key, prompt in child_prompts.items():
new_key = f"{mounted.prefix}_{key}" if mounted.prefix else key
all_prompts[new_key] = prompt.model_copy(key=new_key)
except Exception as e:
logger.warning(
f"Failed to get prompts from mounted server {mounted.server.name!r}: {e}"
)
if fastmcp.settings.mounted_components_raise_on_load_error:
raise
continue
return all_prompts
async def get_prompt(self, key: str) -> Prompt:
prompts = await self.get_prompts()
if key not in prompts:
raise NotFoundError(f"Unknown prompt: {key}")
return prompts[key]
def custom_route(
self,
path: str,
methods: list[str],
name: str | None = None,
include_in_schema: bool = True,
) -> Callable[
[Callable[[Request], Awaitable[Response]]],
Callable[[Request], Awaitable[Response]],
]:
"""
Decorator to register a custom HTTP route on the FastMCP server.
Allows adding arbitrary HTTP endpoints outside the standard MCP protocol,
which can be useful for OAuth callbacks, health checks, or admin APIs.
The handler function must be an async function that accepts a Starlette
Request and returns a Response.
Args:
path: URL path for the route (e.g., "/auth/callback")
methods: List of HTTP methods to support (e.g., ["GET", "POST"])
name: Optional name for the route (to reference this route with
Starlette's reverse URL lookup feature)
include_in_schema: Whether to include in OpenAPI schema, defaults to True
Example:
Register a custom HTTP route for a health check endpoint:
```python
@server.custom_route("/health", methods=["GET"])
async def health_check(request: Request) -> Response:
return JSONResponse({"status": "ok"})
```
"""
def decorator(
fn: Callable[[Request], Awaitable[Response]],
) -> Callable[[Request], Awaitable[Response]]:
self._additional_http_routes.append(
Route(
path,
endpoint=fn,
methods=methods,
name=name,
include_in_schema=include_in_schema,
)
)
return fn
return decorator
def _get_additional_http_routes(self) -> list[BaseRoute]:
"""Get all additional HTTP routes including from mounted servers.
Returns a list of all custom HTTP routes from this server and
recursively from all mounted servers.
Returns:
List of Starlette BaseRoute objects
"""
routes = list(self._additional_http_routes)
# Recursively get routes from mounted servers
for mounted_server in self._mounted_servers:
mounted_routes = mounted_server.server._get_additional_http_routes()
routes.extend(mounted_routes)
return routes
async def _list_tools_mcp(self) -> list[SDKTool]:
"""
List all available tools, in the format expected by the low-level MCP
server.
"""
logger.debug(f"[{self.name}] Handler called: list_tools")
async with fastmcp.server.context.Context(fastmcp=self):
tools = await self._list_tools_middleware()
return [
tool.to_mcp_tool(
name=tool.key,
include_fastmcp_meta=self.include_fastmcp_meta,
)
for tool in tools
]
async def _list_tools_middleware(self) -> list[Tool]:
"""
List all available tools, applying MCP middleware.
"""
async with fastmcp.server.context.Context(fastmcp=self) as fastmcp_ctx:
# Create the middleware context.
mw_context = MiddlewareContext(
message=mcp.types.ListToolsRequest(method="tools/list"),
source="client",
type="request",
method="tools/list",
fastmcp_context=fastmcp_ctx,
)
# Apply the middleware chain.
return list(
await self._apply_middleware(
context=mw_context, call_next=self._list_tools
)
)
async def _list_tools(
self,
context: MiddlewareContext[mcp.types.ListToolsRequest],
) -> list[Tool]:
"""
List all available tools.
"""
# 1. Get local tools and filter them
local_tools = await self._tool_manager.get_tools()
filtered_local = [
tool for tool in local_tools.values() if self._should_enable_component(tool)
]
# 2. Get tools from mounted servers
# Mounted servers apply their own filtering, but we also apply parent's filtering
# Use a dict to implement "later wins" deduplication by key
all_tools: dict[str, Tool] = {tool.key: tool for tool in filtered_local}
for mounted in self._mounted_servers:
try:
child_tools = await mounted.server._list_tools_middleware()
for tool in child_tools:
# Apply parent server's filtering to mounted components
if not self._should_enable_component(tool):
continue
# Check for manual override first, then apply prefix
if mounted.tool_names and tool.key in mounted.tool_names:
key = mounted.tool_names[tool.key]
elif mounted.prefix:
key = f"{mounted.prefix}_{tool.key}"
else:
key = tool.key
if key != tool.key:
tool = tool.model_copy(key=key)
# Later mounted servers override earlier ones
all_tools[key] = tool
except Exception as e:
server_name = getattr(
getattr(mounted, "server", None), "name", repr(mounted)
)
logger.warning(
f"Failed to list tools from mounted server {server_name!r}: {e}"
)
if fastmcp.settings.mounted_components_raise_on_load_error:
raise
continue
return list(all_tools.values())
async def _list_resources_mcp(self) -> list[SDKResource]:
"""
List all available resources, in the format expected by the low-level MCP
server.
"""
logger.debug(f"[{self.name}] Handler called: list_resources")
async with fastmcp.server.context.Context(fastmcp=self):
resources = await self._list_resources_middleware()
return [
resource.to_mcp_resource(
uri=resource.key,
include_fastmcp_meta=self.include_fastmcp_meta,
)
for resource in resources
]
async def _list_resources_middleware(self) -> list[Resource]:
"""
List all available resources, applying MCP middleware.
"""
async with fastmcp.server.context.Context(fastmcp=self) as fastmcp_ctx:
# Create the middleware context.
mw_context = MiddlewareContext(
message={}, # List resources doesn't have parameters
source="client",
type="request",
method="resources/list",
fastmcp_context=fastmcp_ctx,
)
# Apply the middleware chain.
return list(
await self._apply_middleware(
context=mw_context, call_next=self._list_resources
)
)
async def _list_resources(
self,
context: MiddlewareContext[dict[str, Any]],
) -> list[Resource]:
"""
List all available resources.
"""
# 1. Filter local resources
local_resources = await self._resource_manager.get_resources()
filtered_local = [
resource
for resource in local_resources.values()
if self._should_enable_component(resource)
]
# 2. Get from mounted servers with resource prefix handling
# Mounted servers apply their own filtering, but we also apply parent's filtering
# Use a dict to implement "later wins" deduplication by key
all_resources: dict[str, Resource] = {
resource.key: resource for resource in filtered_local
}
for mounted in self._mounted_servers:
try:
child_resources = await mounted.server._list_resources_middleware()
for resource in child_resources:
# Apply parent server's filtering to mounted components
if not self._should_enable_component(resource):
continue
key = resource.key
if mounted.prefix:
key = add_resource_prefix(resource.key, mounted.prefix)
resource = resource.model_copy(
key=key,
update={"name": f"{mounted.prefix}_{resource.name}"},
)
# Later mounted servers override earlier ones
all_resources[key] = resource
except Exception as e:
server_name = getattr(
getattr(mounted, "server", None), "name", repr(mounted)
)
logger.warning(f"Failed to list resources from {server_name!r}: {e}")
if fastmcp.settings.mounted_components_raise_on_load_error:
raise
continue
return list(all_resources.values())
async def _list_resource_templates_mcp(self) -> list[SDKResourceTemplate]:
"""
List all available resource templates, in the format expected by the low-level MCP
server.
"""
logger.debug(f"[{self.name}] Handler called: list_resource_templates")
async with fastmcp.server.context.Context(fastmcp=self):
templates = await self._list_resource_templates_middleware()
return [
template.to_mcp_template(
uriTemplate=template.key,
include_fastmcp_meta=self.include_fastmcp_meta,
)
for template in templates
]
async def _list_resource_templates_middleware(self) -> list[ResourceTemplate]:
"""
List all available resource templates, applying MCP middleware.
"""
async with fastmcp.server.context.Context(fastmcp=self) as fastmcp_ctx:
# Create the middleware context.
mw_context = MiddlewareContext(
message={}, # List resource templates doesn't have parameters
source="client",
type="request",
method="resources/templates/list",
fastmcp_context=fastmcp_ctx,
)
# Apply the middleware chain.
return list(
await self._apply_middleware(
context=mw_context, call_next=self._list_resource_templates
)
)
async def _list_resource_templates(
self,
context: MiddlewareContext[dict[str, Any]],
) -> list[ResourceTemplate]:
"""
List all available resource templates.
"""
# 1. Filter local templates
local_templates = await self._resource_manager.get_resource_templates()
filtered_local = [
template
for template in local_templates.values()
if self._should_enable_component(template)
]
# 2. Get from mounted servers with resource prefix handling
# Mounted servers apply their own filtering, but we also apply parent's filtering
# Use a dict to implement "later wins" deduplication by key
all_templates: dict[str, ResourceTemplate] = {
template.key: template for template in filtered_local
}
for mounted in self._mounted_servers:
try:
child_templates = (
await mounted.server._list_resource_templates_middleware()
)
for template in child_templates:
# Apply parent server's filtering to mounted components
if not self._should_enable_component(template):
continue
key = template.key
if mounted.prefix:
key = add_resource_prefix(template.key, mounted.prefix)
template = template.model_copy(
key=key,
update={"name": f"{mounted.prefix}_{template.name}"},
)
# Later mounted servers override earlier ones
all_templates[key] = template
except Exception as e:
server_name = getattr(
getattr(mounted, "server", None), "name", repr(mounted)
)
logger.warning(
f"Failed to list resource templates from {server_name!r}: {e}"
)
if fastmcp.settings.mounted_components_raise_on_load_error:
raise
continue
return list(all_templates.values())
async def _list_prompts_mcp(self) -> list[SDKPrompt]:
"""
List all available prompts, in the format expected by the low-level MCP
server.
"""
logger.debug(f"[{self.name}] Handler called: list_prompts")
async with fastmcp.server.context.Context(fastmcp=self):
prompts = await self._list_prompts_middleware()
return [
prompt.to_mcp_prompt(
name=prompt.key,
include_fastmcp_meta=self.include_fastmcp_meta,
)
for prompt in prompts
]
async def _list_prompts_middleware(self) -> list[Prompt]:
"""
List all available prompts, applying MCP middleware.
"""
async with fastmcp.server.context.Context(fastmcp=self) as fastmcp_ctx:
# Create the middleware context.
mw_context = MiddlewareContext(
message=mcp.types.ListPromptsRequest(method="prompts/list"),
source="client",
type="request",
method="prompts/list",
fastmcp_context=fastmcp_ctx,
)
# Apply the middleware chain.
return list(
await self._apply_middleware(
context=mw_context, call_next=self._list_prompts
)
)
async def _list_prompts(
self,
context: MiddlewareContext[mcp.types.ListPromptsRequest],
) -> list[Prompt]:
"""
List all available prompts.
"""
# 1. Filter local prompts
local_prompts = await self._prompt_manager.get_prompts()
filtered_local = [
prompt
for prompt in local_prompts.values()
if self._should_enable_component(prompt)
]
# 2. Get from mounted servers
# Mounted servers apply their own filtering, but we also apply parent's filtering
# Use a dict to implement "later wins" deduplication by key
all_prompts: dict[str, Prompt] = {
prompt.key: prompt for prompt in filtered_local
}
for mounted in self._mounted_servers:
try:
child_prompts = await mounted.server._list_prompts_middleware()
for prompt in child_prompts:
# Apply parent server's filtering to mounted components
if not self._should_enable_component(prompt):
continue
# Apply prefix to prompt key
if mounted.prefix:
key = f"{mounted.prefix}_{prompt.key}"
else:
key = prompt.key
if key != prompt.key:
prompt = prompt.model_copy(key=key)
# Later mounted servers override earlier ones
all_prompts[key] = prompt
except Exception as e:
server_name = getattr(
getattr(mounted, "server", None), "name", repr(mounted)
)
logger.warning(
f"Failed to list prompts from mounted server {server_name!r}: {e}"
)
if fastmcp.settings.mounted_components_raise_on_load_error:
raise
continue
return list(all_prompts.values())
async def _call_tool_mcp(
self, key: str, arguments: dict[str, Any]
) -> (
list[ContentBlock]
| tuple[list[ContentBlock], dict[str, Any]]
| mcp.types.CallToolResult
):
"""
Handle MCP 'callTool' requests.
Detects SEP-1686 task metadata and routes to background execution if supported.
Args:
key: The name of the tool to call
arguments: Arguments to pass to the tool
Returns:
List of MCP Content objects containing the tool results
"""
logger.debug(
f"[{self.name}] Handler called: call_tool %s with %s", key, arguments
)
async with fastmcp.server.context.Context(fastmcp=self):
try:
# Check for SEP-1686 task metadata via request context
task_meta = None
try:
# Access task metadata from SDK's request context
ctx = self._mcp_server.request_context
if ctx.experimental.is_task:
task_meta = ctx.experimental.task_metadata
except (AttributeError, LookupError):
# No request context available - proceed without task metadata
pass
# Get tool from local manager, mounted servers, or proxy
tool = await self._get_tool_with_task_config(key)
if (
tool
and self._should_enable_component(tool)
and hasattr(tool, "task_config")
):
task_mode = tool.task_config.mode # type: ignore[union-attr]
# Enforce mode="required" - must have task metadata
if task_mode == "required" and not task_meta:
raise McpError(
ErrorData(
code=METHOD_NOT_FOUND,
message=f"Tool '{key}' requires task-augmented execution",
)
)
# Route to background if task metadata present and mode allows
if task_meta and task_mode != "forbidden":
# For FunctionTool, use Docket for background execution
if isinstance(tool, FunctionTool):
task_meta_dict = task_meta.model_dump(exclude_none=True)
return await handle_tool_as_task(
self, key, arguments, task_meta_dict
)
# For ProxyTool/mounted tools, proceed with normal execution
# They will forward task metadata to their backend
# Forbidden mode: task requested but mode="forbidden"
# Return error result with returned_immediately=True
if task_meta and task_mode == "forbidden":
return mcp.types.CallToolResult(
content=[
mcp.types.TextContent(
type="text",
text=f"Tool '{key}' does not support task-augmented execution",
)
],
isError=True,
_meta={
"modelcontextprotocol.io/task": {
"returned_immediately": True
}
},
)
# Synchronous execution (normal path)
result = await self._call_tool_middleware(key, arguments)
return result.to_mcp_result()
except DisabledError as e:
raise NotFoundError(f"Unknown tool: {key}") from e
except NotFoundError as e:
raise NotFoundError(f"Unknown tool: {key}") from e
async def _call_tool_middleware(
self,
key: str,
arguments: dict[str, Any],
) -> ToolResult:
"""
Applies this server's middleware and delegates the filtered call to the manager.
"""
mw_context = MiddlewareContext[CallToolRequestParams](
message=mcp.types.CallToolRequestParams(name=key, arguments=arguments),
source="client",
type="request",
method="tools/call",
fastmcp_context=fastmcp.server.dependencies.get_context(),
)
return await self._apply_middleware(
context=mw_context, call_next=self._call_tool
)
async def _call_tool(
self,
context: MiddlewareContext[mcp.types.CallToolRequestParams],
) -> ToolResult:
"""
Call a tool
"""
tool_name = context.message.name
# Try mounted servers in reverse order (later wins)
for mounted in reversed(self._mounted_servers):
try_name = tool_name
# First check if tool_name is an overridden name (reverse lookup)
if mounted.tool_names:
for orig_key, override_name in mounted.tool_names.items():
if override_name == tool_name:
try_name = orig_key
break
else:
# Not an override, try standard prefix stripping
if mounted.prefix:
if not tool_name.startswith(f"{mounted.prefix}_"):
continue
try_name = tool_name[len(mounted.prefix) + 1 :]
elif mounted.prefix:
if not tool_name.startswith(f"{mounted.prefix}_"):
continue
try_name = tool_name[len(mounted.prefix) + 1 :]
try:
# First, get the tool to check if parent's filter allows it
# Use get_tool() instead of _tool_manager.get_tool() to support
# nested mounted servers (tools mounted more than 2 levels deep)
tool = await mounted.server.get_tool(try_name)
if not self._should_enable_component(tool):
# Parent filter blocks this tool, continue searching
continue
return await mounted.server._call_tool_middleware(
try_name, context.message.arguments or {}
)
except NotFoundError:
continue
# Try local tools last (mounted servers override local)
try:
tool = await self._tool_manager.get_tool(tool_name)
if self._should_enable_component(tool):
return await self._tool_manager.call_tool(
key=tool_name, arguments=context.message.arguments or {}
)
except NotFoundError:
pass
raise NotFoundError(f"Unknown tool: {tool_name!r}")
async def _read_resource_mcp(self, uri: AnyUrl | str) -> list[ReadResourceContents]:
"""
Handle MCP 'readResource' requests.
Delegates to _read_resource, which should be overridden by FastMCP subclasses.
"""
logger.debug(f"[{self.name}] Handler called: read_resource %s", uri)
async with fastmcp.server.context.Context(fastmcp=self):
try:
# Task routing handled by custom handler
return list[ReadResourceContents](
await self._read_resource_middleware(uri)
)
except DisabledError as e:
# convert to NotFoundError to avoid leaking resource presence
raise NotFoundError(f"Unknown resource: {str(uri)!r}") from e
except NotFoundError as e:
# standardize NotFound message
raise NotFoundError(f"Unknown resource: {str(uri)!r}") from e
async def _read_resource_middleware(
self,
uri: AnyUrl | str,
) -> list[ReadResourceContents]:
"""
Applies this server's middleware and delegates the filtered call to the manager.
"""
# Convert string URI to AnyUrl if needed
uri_param = AnyUrl(uri) if isinstance(uri, str) else uri
mw_context = MiddlewareContext(
message=mcp.types.ReadResourceRequestParams(uri=uri_param),
source="client",
type="request",
method="resources/read",
fastmcp_context=fastmcp.server.dependencies.get_context(),
)
return list(
await self._apply_middleware(
context=mw_context, call_next=self._read_resource
)
)
async def _read_resource(
self,
context: MiddlewareContext[mcp.types.ReadResourceRequestParams],
) -> list[ReadResourceContents]:
"""
Read a resource
"""
uri_str = str(context.message.uri)
# Try mounted servers in reverse order (later wins)
for mounted in reversed(self._mounted_servers):
key = uri_str
if mounted.prefix:
if not has_resource_prefix(key, mounted.prefix):
continue
key = remove_resource_prefix(key, mounted.prefix)
# First, get the resource/template to check if parent's filter allows it
# Use get_resource_or_template to support nested mounted servers
# (resources/templates mounted more than 2 levels deep)
resource = await mounted.server._get_resource_or_template_or_none(key)
if resource is None:
continue
if not self._should_enable_component(resource):
# Parent filter blocks this resource, continue searching
continue
try:
result = list(await mounted.server._read_resource_middleware(key))
return result
except NotFoundError:
continue
# Try local resources last (mounted servers override local)
try:
resource = await self._resource_manager.get_resource(uri_str)
if self._should_enable_component(resource):
content = await self._resource_manager.read_resource(uri_str)
return [
ReadResourceContents(
content=content,
mime_type=resource.mime_type,
)
]
except NotFoundError:
pass
raise NotFoundError(f"Unknown resource: {uri_str!r}")
async def _get_prompt_mcp(
self, name: str, arguments: dict[str, Any] | None = None
) -> GetPromptResult:
"""
Handle MCP 'getPrompt' requests.
Delegates to _get_prompt, which should be overridden by FastMCP subclasses.
"""
import fastmcp.server.context
logger.debug(
f"[{self.name}] Handler called: get_prompt %s with %s", name, arguments
)
async with fastmcp.server.context.Context(fastmcp=self):
try:
# Task routing handled by custom handler
return await self._get_prompt_middleware(name, arguments)
except DisabledError as e:
# convert to NotFoundError to avoid leaking prompt presence
raise NotFoundError(f"Unknown prompt: {name}") from e
except NotFoundError as e:
# standardize NotFound message
raise NotFoundError(f"Unknown prompt: {name}") from e
async def _get_prompt_middleware(
self, name: str, arguments: dict[str, Any] | None = None
) -> GetPromptResult:
"""
Applies this server's middleware and delegates the filtered call to the manager.
"""
mw_context = MiddlewareContext(
message=mcp.types.GetPromptRequestParams(name=name, arguments=arguments),
source="client",
type="request",
method="prompts/get",
fastmcp_context=fastmcp.server.dependencies.get_context(),
)
return await self._apply_middleware(
context=mw_context, call_next=self._get_prompt
)
async def _get_prompt(
self,
context: MiddlewareContext[mcp.types.GetPromptRequestParams],
) -> GetPromptResult:
name = context.message.name
# Try mounted servers in reverse order (later wins)
for mounted in reversed(self._mounted_servers):
try_name = name
if mounted.prefix:
if not name.startswith(f"{mounted.prefix}_"):
continue
try_name = name[len(mounted.prefix) + 1 :]
try:
# First, get the prompt to check if parent's filter allows it
# Use get_prompt() instead of _prompt_manager.get_prompt() to support
# nested mounted servers (prompts mounted more than 2 levels deep)
prompt = await mounted.server.get_prompt(try_name)
if not self._should_enable_component(prompt):
# Parent filter blocks this prompt, continue searching
continue
return await mounted.server._get_prompt_middleware(
try_name, context.message.arguments
)
except NotFoundError:
continue
# Try local prompts last (mounted servers override local)
try:
prompt = await self._prompt_manager.get_prompt(name)
if self._should_enable_component(prompt):
return await self._prompt_manager.render_prompt(
name=name, arguments=context.message.arguments
)
except NotFoundError:
pass
raise NotFoundError(f"Unknown prompt: {name!r}")
def add_tool(self, tool: Tool) -> Tool:
"""Add a tool to the server.
The tool function can optionally request a Context object by adding a parameter
with the Context type annotation. See the @tool decorator for examples.
Args:
tool: The Tool instance to register
Returns:
The tool instance that was added to the server.
"""
self._tool_manager.add_tool(tool)
# Send notification if we're in a request context
try:
from fastmcp.server.dependencies import get_context
context = get_context()
context._queue_tool_list_changed() # type: ignore[private-use]
except RuntimeError:
pass # No context available
return tool
def remove_tool(self, name: str) -> None:
"""Remove a tool from the server.
Args:
name: The name of the tool to remove
Raises:
NotFoundError: If the tool is not found
"""
self._tool_manager.remove_tool(name)
# Send notification if we're in a request context
try:
from fastmcp.server.dependencies import get_context
context = get_context()
context._queue_tool_list_changed() # type: ignore[private-use]
except RuntimeError:
pass # No context available
def add_tool_transformation(
self, tool_name: str, transformation: ToolTransformConfig
) -> None:
"""Add a tool transformation."""
self._tool_manager.add_tool_transformation(tool_name, transformation)
def remove_tool_transformation(self, tool_name: str) -> None:
"""Remove a tool transformation."""
self._tool_manager.remove_tool_transformation(tool_name)
@overload
def tool(
self,
name_or_fn: AnyFunction,
*,
name: str | None = None,
title: str | None = None,
description: str | None = None,
icons: list[mcp.types.Icon] | None = None,
tags: set[str] | None = None,
output_schema: dict[str, Any] | NotSetT | None = NotSet,
annotations: ToolAnnotations | dict[str, Any] | None = None,
exclude_args: list[str] | None = None,
meta: dict[str, Any] | None = None,
enabled: bool | None = None,
task: bool | TaskConfig | None = None,
) -> FunctionTool: ...
@overload
def tool(
self,
name_or_fn: str | None = None,
*,
name: str | None = None,
title: str | None = None,
description: str | None = None,
icons: list[mcp.types.Icon] | None = None,
tags: set[str] | None = None,
output_schema: dict[str, Any] | NotSetT | None = NotSet,
annotations: ToolAnnotations | dict[str, Any] | None = None,
exclude_args: list[str] | None = None,
meta: dict[str, Any] | None = None,
enabled: bool | None = None,
task: bool | TaskConfig | None = None,
) -> Callable[[AnyFunction], FunctionTool]: ...
def tool(
self,
name_or_fn: str | AnyFunction | None = None,
*,
name: str | None = None,
title: str | None = None,
description: str | None = None,
icons: list[mcp.types.Icon] | None = None,
tags: set[str] | None = None,
output_schema: dict[str, Any] | NotSetT | None = NotSet,
annotations: ToolAnnotations | dict[str, Any] | None = None,
exclude_args: list[str] | None = None,
meta: dict[str, Any] | None = None,
enabled: bool | None = None,
task: bool | TaskConfig | None = None,
) -> Callable[[AnyFunction], FunctionTool] | FunctionTool:
"""Decorator to register a tool.
Tools can optionally request a Context object by adding a parameter with the
Context type annotation. The context provides access to MCP capabilities like
logging, progress reporting, and resource access.
This decorator supports multiple calling patterns:
- @server.tool (without parentheses)
- @server.tool (with empty parentheses)
- @server.tool("custom_name") (with name as first argument)
- @server.tool(name="custom_name") (with name as keyword argument)
- server.tool(function, name="custom_name") (direct function call)
Args:
name_or_fn: Either a function (when used as @tool), a string name, or None
name: Optional name for the tool (keyword-only, alternative to name_or_fn)
description: Optional description of what the tool does
tags: Optional set of tags for categorizing the tool
output_schema: Optional JSON schema for the tool's output
annotations: Optional annotations about the tool's behavior
exclude_args: Optional list of argument names to exclude from the tool schema.
Deprecated: Use `Depends()` for dependency injection instead.
meta: Optional meta information about the tool
enabled: Optional boolean to enable or disable the tool
Examples:
Register a tool with a custom name:
```python
@server.tool
def my_tool(x: int) -> str:
return str(x)
# Register a tool with a custom name
@server.tool
def my_tool(x: int) -> str:
return str(x)
@server.tool("custom_name")
def my_tool(x: int) -> str:
return str(x)
@server.tool(name="custom_name")
def my_tool(x: int) -> str:
return str(x)
# Direct function call
server.tool(my_function, name="custom_name")
```
"""
if isinstance(annotations, dict):
annotations = ToolAnnotations(**annotations)
if isinstance(name_or_fn, classmethod):
raise ValueError(
inspect.cleandoc(
"""
To decorate a classmethod, first define the method and then call
tool() directly on the method instead of using it as a
decorator. See https://gofastmcp.com/patterns/decorating-methods
for examples and more information.
"""
)
)
# Determine the actual name and function based on the calling pattern
if inspect.isroutine(name_or_fn):
# Case 1: @tool (without parens) - function passed directly
# Case 2: direct call like tool(fn, name="something")
fn = name_or_fn
tool_name = name # Use keyword name if provided, otherwise None
# Resolve task parameter
supports_task: bool | TaskConfig = (
task if task is not None else self._support_tasks_by_default
)
# Register the tool immediately and return the tool object
# Note: Deprecation warning for exclude_args is handled in Tool.from_function
tool = Tool.from_function(
fn,
name=tool_name,
title=title,
description=description,
icons=icons,
tags=tags,
output_schema=output_schema,
annotations=annotations,
exclude_args=exclude_args,
meta=meta,
serializer=self._tool_serializer,
enabled=enabled,
task=supports_task,
)
self.add_tool(tool)
return tool
elif isinstance(name_or_fn, str):
# Case 3: @tool("custom_name") - name passed as first argument
if name is not None:
raise TypeError(
"Cannot specify both a name as first argument and as keyword argument. "
f"Use either @tool('{name_or_fn}') or @tool(name='{name}'), not both."
)
tool_name = name_or_fn
elif name_or_fn is None:
# Case 4: @tool or @tool(name="something") - use keyword name
tool_name = name
else:
raise TypeError(
f"First argument to @tool must be a function, string, or None, got {type(name_or_fn)}"
)
# Return partial for cases where we need to wait for the function
return partial(
self.tool,
name=tool_name,
title=title,
description=description,
icons=icons,
tags=tags,
output_schema=output_schema,
annotations=annotations,
exclude_args=exclude_args,
meta=meta,
enabled=enabled,
task=task,
)
def add_resource(self, resource: Resource) -> Resource:
"""Add a resource to the server.
Args:
resource: A Resource instance to add
Returns:
The resource instance that was added to the server.
"""
self._resource_manager.add_resource(resource)
# Send notification if we're in a request context
try:
from fastmcp.server.dependencies import get_context
context = get_context()
context._queue_resource_list_changed() # type: ignore[private-use]
except RuntimeError:
pass # No context available
return resource
def add_template(self, template: ResourceTemplate) -> ResourceTemplate:
"""Add a resource template to the server.
Args:
template: A ResourceTemplate instance to add
Returns:
The template instance that was added to the server.
"""
self._resource_manager.add_template(template)
# Send notification if we're in a request context
try:
from fastmcp.server.dependencies import get_context
context = get_context()
context._queue_resource_list_changed() # type: ignore[private-use]
except RuntimeError:
pass # No context available
return template
def resource(
self,
uri: str,
*,
name: str | None = None,
title: str | None = None,
description: str | None = None,
icons: list[mcp.types.Icon] | None = None,
mime_type: str | None = None,
tags: set[str] | None = None,
enabled: bool | None = None,
annotations: Annotations | dict[str, Any] | None = None,
meta: dict[str, Any] | None = None,
task: bool | TaskConfig | None = None,
) -> Callable[[AnyFunction], Resource | ResourceTemplate]:
"""Decorator to register a function as a resource.
The function will be called when the resource is read to generate its content.
The function can return:
- str for text content
- bytes for binary content
- other types will be converted to JSON
Resources can optionally request a Context object by adding a parameter with the
Context type annotation. The context provides access to MCP capabilities like
logging, progress reporting, and session information.
If the URI contains parameters (e.g. "resource://{param}") or the function
has parameters, it will be registered as a template resource.
Args:
uri: URI for the resource (e.g. "resource://my-resource" or "resource://{param}")
name: Optional name for the resource
description: Optional description of the resource
mime_type: Optional MIME type for the resource
tags: Optional set of tags for categorizing the resource
enabled: Optional boolean to enable or disable the resource
annotations: Optional annotations about the resource's behavior
meta: Optional meta information about the resource
Examples:
Register a resource with a custom name:
```python
@server.resource("resource://my-resource")
def get_data() -> str:
return "Hello, world!"
@server.resource("resource://my-resource")
async get_data() -> str:
data = await fetch_data()
return f"Hello, world! {data}"
@server.resource("resource://{city}/weather")
def get_weather(city: str) -> str:
return f"Weather for {city}"
@server.resource("resource://{city}/weather")
async def get_weather_with_context(city: str, ctx: Context) -> str:
await ctx.info(f"Fetching weather for {city}")
return f"Weather for {city}"
@server.resource("resource://{city}/weather")
async def get_weather(city: str) -> str:
data = await fetch_weather(city)
return f"Weather for {city}: {data}"
```
"""
if isinstance(annotations, dict):
annotations = Annotations(**annotations)
# Check if user passed function directly instead of calling decorator
if inspect.isroutine(uri):
raise TypeError(
"The @resource decorator was used incorrectly. "
"Did you forget to call it? Use @resource('uri') instead of @resource"
)
def decorator(fn: AnyFunction) -> Resource | ResourceTemplate:
if isinstance(fn, classmethod): # type: ignore[reportUnnecessaryIsInstance]
raise ValueError(
inspect.cleandoc(
"""
To decorate a classmethod, first define the method and then call
resource() directly on the method instead of using it as a
decorator. See https://gofastmcp.com/patterns/decorating-methods
for examples and more information.
"""
)
)
# Resolve task parameter
supports_task: bool | TaskConfig = (
task if task is not None else self._support_tasks_by_default
)
# Check if this should be a template
has_uri_params = "{" in uri and "}" in uri
# Use wrapper to check for user-facing parameters
from fastmcp.server.dependencies import without_injected_parameters
wrapper_fn = without_injected_parameters(fn)
has_func_params = bool(inspect.signature(wrapper_fn).parameters)
if has_uri_params or has_func_params:
template = ResourceTemplate.from_function(
fn=fn,
uri_template=uri,
name=name,
title=title,
description=description,
icons=icons,
mime_type=mime_type,
tags=tags,
enabled=enabled,
annotations=annotations,
meta=meta,
task=supports_task,
)
self.add_template(template)
return template
elif not has_uri_params and not has_func_params:
resource = Resource.from_function(
fn=fn,
uri=uri,
name=name,
title=title,
description=description,
icons=icons,
mime_type=mime_type,
tags=tags,
enabled=enabled,
annotations=annotations,
meta=meta,
task=supports_task,
)
self.add_resource(resource)
return resource
else:
raise ValueError(
"Invalid resource or template definition due to a "
"mismatch between URI parameters and function parameters."
)
return decorator
def add_prompt(self, prompt: Prompt) -> Prompt:
"""Add a prompt to the server.
Args:
prompt: A Prompt instance to add
Returns:
The prompt instance that was added to the server.
"""
self._prompt_manager.add_prompt(prompt)
# Send notification if we're in a request context
try:
from fastmcp.server.dependencies import get_context
context = get_context()
context._queue_prompt_list_changed() # type: ignore[private-use]
except RuntimeError:
pass # No context available
return prompt
@overload
def prompt(
self,
name_or_fn: AnyFunction,
*,
name: str | None = None,
title: str | None = None,
description: str | None = None,
icons: list[mcp.types.Icon] | None = None,
tags: set[str] | None = None,
enabled: bool | None = None,
meta: dict[str, Any] | None = None,
task: bool | TaskConfig | None = None,
) -> FunctionPrompt: ...
@overload
def prompt(
self,
name_or_fn: str | None = None,
*,
name: str | None = None,
title: str | None = None,
description: str | None = None,
icons: list[mcp.types.Icon] | None = None,
tags: set[str] | None = None,
enabled: bool | None = None,
meta: dict[str, Any] | None = None,
task: bool | TaskConfig | None = None,
) -> Callable[[AnyFunction], FunctionPrompt]: ...
def prompt(
self,
name_or_fn: str | AnyFunction | None = None,
*,
name: str | None = None,
title: str | None = None,
description: str | None = None,
icons: list[mcp.types.Icon] | None = None,
tags: set[str] | None = None,
enabled: bool | None = None,
meta: dict[str, Any] | None = None,
task: bool | TaskConfig | None = None,
) -> Callable[[AnyFunction], FunctionPrompt] | FunctionPrompt:
"""Decorator to register a prompt.
Prompts can optionally request a Context object by adding a parameter with the
Context type annotation. The context provides access to MCP capabilities like
logging, progress reporting, and session information.
This decorator supports multiple calling patterns:
- @server.prompt (without parentheses)
- @server.prompt() (with empty parentheses)
- @server.prompt("custom_name") (with name as first argument)
- @server.prompt(name="custom_name") (with name as keyword argument)
- server.prompt(function, name="custom_name") (direct function call)
Args:
name_or_fn: Either a function (when used as @prompt), a string name, or None
name: Optional name for the prompt (keyword-only, alternative to name_or_fn)
description: Optional description of what the prompt does
tags: Optional set of tags for categorizing the prompt
enabled: Optional boolean to enable or disable the prompt
meta: Optional meta information about the prompt
Examples:
```python
@server.prompt
def analyze_table(table_name: str) -> list[Message]:
schema = read_table_schema(table_name)
return [
{
"role": "user",
"content": f"Analyze this schema:\n{schema}"
}
]
@server.prompt()
async def analyze_with_context(table_name: str, ctx: Context) -> list[Message]:
await ctx.info(f"Analyzing table {table_name}")
schema = read_table_schema(table_name)
return [
{
"role": "user",
"content": f"Analyze this schema:\n{schema}"
}
]
@server.prompt("custom_name")
async def analyze_file(path: str) -> list[Message]:
content = await read_file(path)
return [
{
"role": "user",
"content": {
"type": "resource",
"resource": {
"uri": f"file://{path}",
"text": content
}
}
}
]
@server.prompt(name="custom_name")
def another_prompt(data: str) -> list[Message]:
return [{"role": "user", "content": data}]
# Direct function call
server.prompt(my_function, name="custom_name")
```
"""
if isinstance(name_or_fn, classmethod):
raise ValueError(
inspect.cleandoc(
"""
To decorate a classmethod, first define the method and then call
prompt() directly on the method instead of using it as a
decorator. See https://gofastmcp.com/patterns/decorating-methods
for examples and more information.
"""
)
)
# Determine the actual name and function based on the calling pattern
if inspect.isroutine(name_or_fn):
# Case 1: @prompt (without parens) - function passed directly as decorator
# Case 2: direct call like prompt(fn, name="something")
fn = name_or_fn
prompt_name = name # Use keyword name if provided, otherwise None
# Resolve task parameter
supports_task: bool | TaskConfig = (
task if task is not None else self._support_tasks_by_default
)
# Register the prompt immediately
prompt = Prompt.from_function(
fn=fn,
name=prompt_name,
title=title,
description=description,
icons=icons,
tags=tags,
enabled=enabled,
meta=meta,
task=supports_task,
)
self.add_prompt(prompt)
return prompt
elif isinstance(name_or_fn, str):
# Case 3: @prompt("custom_name") - name passed as first argument
if name is not None:
raise TypeError(
"Cannot specify both a name as first argument and as keyword argument. "
f"Use either @prompt('{name_or_fn}') or @prompt(name='{name}'), not both."
)
prompt_name = name_or_fn
elif name_or_fn is None:
# Case 4: @prompt() or @prompt(name="something") - use keyword name
prompt_name = name
else:
raise TypeError(
f"First argument to @prompt must be a function, string, or None, got {type(name_or_fn)}"
)
# Return partial for cases where we need to wait for the function
return partial(
self.prompt,
name=prompt_name,
title=title,
description=description,
icons=icons,
tags=tags,
enabled=enabled,
meta=meta,
task=task,
)
async def run_stdio_async(
self, show_banner: bool = True, log_level: str | None = None
) -> None:
"""Run the server using stdio transport.
Args:
show_banner: Whether to display the server banner
log_level: Log level for the server
"""
# Display server banner
if show_banner:
log_server_banner(server=self)
with temporary_log_level(log_level):
async with self._lifespan_manager():
async with stdio_server() as (read_stream, write_stream):
logger.info(
f"Starting MCP server {self.name!r} with transport 'stdio'"
)
# Build experimental capabilities
experimental_capabilities = get_task_capabilities()
await self._mcp_server.run(
read_stream,
write_stream,
self._mcp_server.create_initialization_options(
notification_options=NotificationOptions(
tools_changed=True
),
experimental_capabilities=experimental_capabilities,
),
)
async def run_http_async(
self,
show_banner: bool = True,
transport: Literal["http", "streamable-http", "sse"] = "http",
host: str | None = None,
port: int | None = None,
log_level: str | None = None,
path: str | None = None,
uvicorn_config: dict[str, Any] | None = None,
middleware: list[ASGIMiddleware] | None = None,
json_response: bool | None = None,
stateless_http: bool | None = None,
) -> None:
"""Run the server using HTTP transport.
Args:
transport: Transport protocol to use - either "streamable-http" (default) or "sse"
host: Host address to bind to (defaults to settings.host)
port: Port to bind to (defaults to settings.port)
log_level: Log level for the server (defaults to settings.log_level)
path: Path for the endpoint (defaults to settings.streamable_http_path or settings.sse_path)
uvicorn_config: Additional configuration for the Uvicorn server
middleware: A list of middleware to apply to the app
json_response: Whether to use JSON response format (defaults to settings.json_response)
stateless_http: Whether to use stateless HTTP (defaults to settings.stateless_http)
"""
host = host or self._deprecated_settings.host
port = port or self._deprecated_settings.port
default_log_level_to_use = (
log_level or self._deprecated_settings.log_level
).lower()
app = self.http_app(
path=path,
transport=transport,
middleware=middleware,
json_response=json_response,
stateless_http=stateless_http,
)
# Display server banner
if show_banner:
log_server_banner(server=self)
uvicorn_config_from_user = uvicorn_config or {}
config_kwargs: dict[str, Any] = {
"timeout_graceful_shutdown": 0,
"lifespan": "on",
"ws": "websockets-sansio",
}
config_kwargs.update(uvicorn_config_from_user)
if "log_config" not in config_kwargs and "log_level" not in config_kwargs:
config_kwargs["log_level"] = default_log_level_to_use
with temporary_log_level(log_level):
async with self._lifespan_manager():
config = uvicorn.Config(app, host=host, port=port, **config_kwargs)
server = uvicorn.Server(config)
path = app.state.path.lstrip("/") # type: ignore
logger.info(
f"Starting MCP server {self.name!r} with transport {transport!r} on http://{host}:{port}/{path}"
)
await server.serve()
def http_app(
self,
path: str | None = None,
middleware: list[ASGIMiddleware] | None = None,
json_response: bool | None = None,
stateless_http: bool | None = None,
transport: Literal["http", "streamable-http", "sse"] = "http",
event_store: EventStore | None = None,
retry_interval: int | None = None,
) -> StarletteWithLifespan:
"""Create a Starlette app using the specified HTTP transport.
Args:
path: The path for the HTTP endpoint
middleware: A list of middleware to apply to the app
json_response: Whether to use JSON response format
stateless_http: Whether to use stateless mode (new transport per request)
transport: Transport protocol to use - "http", "streamable-http", or "sse"
event_store: Optional event store for SSE polling/resumability. When set,
enables clients to reconnect and resume receiving events after
server-initiated disconnections. Only used with streamable-http transport.
retry_interval: Optional retry interval in milliseconds for SSE polling.
Controls how quickly clients should reconnect after server-initiated
disconnections. Requires event_store to be set. Only used with
streamable-http transport.
Returns:
A Starlette application configured with the specified transport
"""
if transport in ("streamable-http", "http"):
return create_streamable_http_app(
server=self,
streamable_http_path=path
or self._deprecated_settings.streamable_http_path,
event_store=event_store,
retry_interval=retry_interval,
auth=self.auth,
json_response=(
json_response
if json_response is not None
else self._deprecated_settings.json_response
),
stateless_http=(
stateless_http
if stateless_http is not None
else self._deprecated_settings.stateless_http
),
debug=self._deprecated_settings.debug,
middleware=middleware,
)
elif transport == "sse":
return create_sse_app(
server=self,
message_path=self._deprecated_settings.message_path,
sse_path=path or self._deprecated_settings.sse_path,
auth=self.auth,
debug=self._deprecated_settings.debug,
middleware=middleware,
)
def mount(
self,
server: FastMCP[LifespanResultT],
prefix: str | None = None,
as_proxy: bool | None = None,
tool_names: dict[str, str] | None = None,
) -> None:
"""Mount another FastMCP server on this server with an optional prefix.
Unlike importing (with import_server), mounting establishes a dynamic connection
between servers. When a client interacts with a mounted server's objects through
the parent server, requests are forwarded to the mounted server in real-time.
This means changes to the mounted server are immediately reflected when accessed
through the parent.
When a server is mounted with a prefix:
- Tools from the mounted server are accessible with prefixed names.
Example: If server has a tool named "get_weather", it will be available as "prefix_get_weather".
- Resources are accessible with prefixed URIs.
Example: If server has a resource with URI "weather://forecast", it will be available as
"weather://prefix/forecast".
- Templates are accessible with prefixed URI templates.
Example: If server has a template with URI "weather://location/{id}", it will be available
as "weather://prefix/location/{id}".
- Prompts are accessible with prefixed names.
Example: If server has a prompt named "weather_prompt", it will be available as
"prefix_weather_prompt".
When a server is mounted without a prefix (prefix=None), its tools, resources, templates,
and prompts are accessible with their original names. Multiple servers can be mounted
without prefixes, and they will be tried in order until a match is found.
There are two modes for mounting servers:
1. Direct mounting (default when server has no custom lifespan): The parent server
directly accesses the mounted server's objects in-memory for better performance.
In this mode, no client lifecycle events occur on the mounted server, including
lifespan execution.
2. Proxy mounting (default when server has a custom lifespan): The parent server
treats the mounted server as a separate entity and communicates with it via a
Client transport. This preserves all client-facing behaviors, including lifespan
execution, but with slightly higher overhead.
Args:
server: The FastMCP server to mount.
prefix: Optional prefix to use for the mounted server's objects. If None,
the server's objects are accessible with their original names.
as_proxy: Whether to treat the mounted server as a proxy. If None (default),
automatically determined based on whether the server has a custom lifespan
(True if it has a custom lifespan, False otherwise).
tool_names: Optional mapping of original tool names to custom names. Use this
to override prefixed names. Keys are the original tool names from the
mounted server.
"""
from fastmcp.server.proxy import FastMCPProxy
# if as_proxy is not specified and the server has a custom lifespan,
# we should treat it as a proxy
if as_proxy is None:
as_proxy = server._lifespan != default_lifespan
if as_proxy and not isinstance(server, FastMCPProxy):
server = FastMCP.as_proxy(server)
# Mark the server as mounted so it skips creating its own Docket/Worker.
# The parent's Docket handles task execution, avoiding race conditions
# with multiple workers competing for tasks from the same queue.
server._is_mounted = True
# Delegate mounting to all three managers
mounted_server = MountedServer(
prefix=prefix,
server=server,
tool_names=tool_names,
)
self._mounted_servers.append(mounted_server)
async def import_server(
self,
server: FastMCP[LifespanResultT],
prefix: str | None = None,
) -> None:
"""
Import the MCP objects from another FastMCP server into this one,
optionally with a given prefix.
Note that when a server is *imported*, its objects are immediately
registered to the importing server. This is a one-time operation and
future changes to the imported server will not be reflected in the
importing server. Server-level configurations and lifespans are not imported.
When a server is imported with a prefix:
- The tools are imported with prefixed names
Example: If server has a tool named "get_weather", it will be
available as "prefix_get_weather"
- The resources are imported with prefixed URIs using the new format
Example: If server has a resource with URI "weather://forecast", it will
be available as "weather://prefix/forecast"
- The templates are imported with prefixed URI templates using the new format
Example: If server has a template with URI "weather://location/{id}", it will
be available as "weather://prefix/location/{id}"
- The prompts are imported with prefixed names
Example: If server has a prompt named "weather_prompt", it will be available as
"prefix_weather_prompt"
When a server is imported without a prefix (prefix=None), its tools, resources,
templates, and prompts are imported with their original names.
Args:
server: The FastMCP server to import
prefix: Optional prefix to use for the imported server's objects. If None,
objects are imported with their original names.
"""
# Import tools from the server
for key, tool in (await server.get_tools()).items():
if prefix:
tool = tool.model_copy(key=f"{prefix}_{key}")
self._tool_manager.add_tool(tool)
# Import resources and templates from the server
for key, resource in (await server.get_resources()).items():
if prefix:
resource_key = add_resource_prefix(key, prefix)
resource = resource.model_copy(
update={"name": f"{prefix}_{resource.name}"}, key=resource_key
)
self._resource_manager.add_resource(resource)
for key, template in (await server.get_resource_templates()).items():
if prefix:
template_key = add_resource_prefix(key, prefix)
template = template.model_copy(
update={"name": f"{prefix}_{template.name}"}, key=template_key
)
self._resource_manager.add_template(template)
# Import prompts from the server
for key, prompt in (await server.get_prompts()).items():
if prefix:
prompt = prompt.model_copy(key=f"{prefix}_{key}")
self._prompt_manager.add_prompt(prompt)
if server._lifespan != default_lifespan:
from warnings import warn
warn(
message="When importing from a server with a lifespan, the lifespan from the imported server will not be used.",
category=RuntimeWarning,
stacklevel=2,
)
if prefix:
logger.debug(
f"[{self.name}] Imported server {server.name} with prefix '{prefix}'"
)
else:
logger.debug(f"[{self.name}] Imported server {server.name}")
@classmethod
def from_openapi(
cls,
openapi_spec: dict[str, Any],
client: httpx.AsyncClient,
route_maps: list[RouteMap] | None = None,
route_map_fn: OpenAPIRouteMapFn | None = None,
mcp_component_fn: OpenAPIComponentFn | None = None,
mcp_names: dict[str, str] | None = None,
tags: set[str] | None = None,
**settings: Any,
) -> FastMCPOpenAPI:
"""
Create a FastMCP server from an OpenAPI specification.
"""
from .openapi import FastMCPOpenAPI
return FastMCPOpenAPI(
openapi_spec=openapi_spec,
client=client,
route_maps=route_maps,
route_map_fn=route_map_fn,
mcp_component_fn=mcp_component_fn,
mcp_names=mcp_names,
tags=tags,
**settings,
)
@classmethod
def from_fastapi(
cls,
app: Any,
name: str | None = None,
route_maps: list[RouteMap] | None = None,
route_map_fn: OpenAPIRouteMapFn | None = None,
mcp_component_fn: OpenAPIComponentFn | None = None,
mcp_names: dict[str, str] | None = None,
httpx_client_kwargs: dict[str, Any] | None = None,
tags: set[str] | None = None,
**settings: Any,
) -> FastMCPOpenAPI:
"""
Create a FastMCP server from a FastAPI application.
"""
from .openapi import FastMCPOpenAPI
if httpx_client_kwargs is None:
httpx_client_kwargs = {}
httpx_client_kwargs.setdefault("base_url", "http://fastapi")
client = httpx.AsyncClient(
transport=httpx.ASGITransport(app=app),
**httpx_client_kwargs,
)
name = name or app.title
return FastMCPOpenAPI(
openapi_spec=app.openapi(),
client=client,
name=name,
route_maps=route_maps,
route_map_fn=route_map_fn,
mcp_component_fn=mcp_component_fn,
mcp_names=mcp_names,
tags=tags,
**settings,
)
@classmethod
def as_proxy(
cls,
backend: (
Client[ClientTransportT]
| ClientTransport
| FastMCP[Any]
| FastMCP1Server
| AnyUrl
| Path
| MCPConfig
| dict[str, Any]
| str
),
**settings: Any,
) -> FastMCPProxy:
"""Create a FastMCP proxy server for the given backend.
The `backend` argument can be either an existing `fastmcp.client.Client`
instance or any value accepted as the `transport` argument of
`fastmcp.client.Client`. This mirrors the convenience of the
`fastmcp.client.Client` constructor.
"""
from fastmcp.client.client import Client
from fastmcp.server.proxy import FastMCPProxy, ProxyClient
if isinstance(backend, Client):
client = backend
# Session strategy based on client connection state:
# - Connected clients: reuse existing session for all requests
# - Disconnected clients: create fresh sessions per request for isolation
if client.is_connected():
proxy_logger = get_logger(__name__)
proxy_logger.info(
"Proxy detected connected client - reusing existing session for all requests. "
"This may cause context mixing in concurrent scenarios."
)
# Reuse sessions - return the same client instance
def reuse_client_factory():
return client
client_factory = reuse_client_factory
else:
# Fresh sessions per request
def fresh_client_factory():
return client.new()
client_factory = fresh_client_factory
else:
base_client = ProxyClient(backend) # type: ignore
# Fresh client created from transport - use fresh sessions per request
def proxy_client_factory():
return base_client.new()
client_factory = proxy_client_factory
return FastMCPProxy(client_factory=client_factory, **settings)
def _should_enable_component(
self,
component: FastMCPComponent,
) -> bool:
"""
Given a component, determine if it should be enabled. Returns True if it should be enabled; False if it should not.
Rules:
- If the component's enabled property is False, always return False.
- If both include_tags and exclude_tags are None, return True.
- If exclude_tags is provided, check each exclude tag:
- If the exclude tag is a string, it must be present in the input tags to exclude.
- If include_tags is provided, check each include tag:
- If the include tag is a string, it must be present in the input tags to include.
- If include_tags is provided and none of the include tags match, return False.
- If include_tags is not provided, return True.
"""
if not component.enabled:
return False
if self.include_tags is None and self.exclude_tags is None:
return True
if self.exclude_tags is not None:
if any(etag in component.tags for etag in self.exclude_tags):
return False
if self.include_tags is not None:
return bool(any(itag in component.tags for itag in self.include_tags))
return True
@classmethod
def generate_name(cls, name: str | None = None) -> str:
class_name = cls.__name__
if name is None:
return f"{class_name}-{secrets.token_hex(2)}"
else:
return f"{class_name}-{name}-{secrets.token_hex(2)}"
@dataclass
class MountedServer:
prefix: str | None
server: FastMCP[Any]
tool_names: dict[str, str] | None = None
def add_resource_prefix(uri: str, prefix: str) -> str:
"""Add a prefix to a resource URI using path formatting (resource://prefix/path).
Args:
uri: The original resource URI
prefix: The prefix to add
Returns:
The resource URI with the prefix added
Examples:
```python
add_resource_prefix("resource://path/to/resource", "prefix")
"resource://prefix/path/to/resource"
```
With absolute path:
```python
add_resource_prefix("resource:///absolute/path", "prefix")
"resource://prefix//absolute/path"
```
Raises:
ValueError: If the URI doesn't match the expected protocol://path format
"""
if not prefix:
return uri
# Split the URI into protocol and path
match = URI_PATTERN.match(uri)
if not match:
raise ValueError(f"Invalid URI format: {uri}. Expected protocol://path format.")
protocol, path = match.groups()
# Add the prefix to the path
return f"{protocol}{prefix}/{path}"
def remove_resource_prefix(uri: str, prefix: str) -> str:
"""Remove a prefix from a resource URI.
Args:
uri: The resource URI with a prefix
prefix: The prefix to remove
Returns:
The resource URI with the prefix removed
Examples:
```python
remove_resource_prefix("resource://prefix/path/to/resource", "prefix")
"resource://path/to/resource"
```
With absolute path:
```python
remove_resource_prefix("resource://prefix//absolute/path", "prefix")
"resource:///absolute/path"
```
Raises:
ValueError: If the URI doesn't match the expected protocol://path format
"""
if not prefix:
return uri
# Split the URI into protocol and path
match = URI_PATTERN.match(uri)
if not match:
raise ValueError(f"Invalid URI format: {uri}. Expected protocol://path format.")
protocol, path = match.groups()
# Check if the path starts with the prefix followed by a /
prefix_pattern = f"^{re.escape(prefix)}/(.*?)$"
path_match = re.match(prefix_pattern, path)
if not path_match:
return uri
# Return the URI without the prefix
return f"{protocol}{path_match.group(1)}"
def has_resource_prefix(uri: str, prefix: str) -> bool:
"""Check if a resource URI has a specific prefix.
Args:
uri: The resource URI to check
prefix: The prefix to look for
Returns:
True if the URI has the specified prefix, False otherwise
Examples:
```python
has_resource_prefix("resource://prefix/path/to/resource", "prefix")
True
```
With other path:
```python
has_resource_prefix("resource://other/path/to/resource", "prefix")
False
```
Raises:
ValueError: If the URI doesn't match the expected protocol://path format
"""
if not prefix:
return False
# Split the URI into protocol and path
match = URI_PATTERN.match(uri)
if not match:
raise ValueError(f"Invalid URI format: {uri}. Expected protocol://path format.")
_, path = match.groups()
# Check if the path starts with the prefix followed by a /
prefix_pattern = f"^{re.escape(prefix)}/"
return bool(re.match(prefix_pattern, path))
