Skip to content

πŸš€ The fast, Pythonic way to build MCP servers and clients.

Relay the client's sampling result unchanged through proxies (#5747)

Aarnav Noble19c349f74641
diff --git a/fastmcp_slim/fastmcp/server/providers/proxy.py b/fastmcp_slim/fastmcp/server/providers/proxy.py
index 6a1136af7..c923a2ae1 100644
--- a/fastmcp_slim/fastmcp/server/providers/proxy.py
+++ b/fastmcp_slim/fastmcp/server/providers/proxy.py
@@ -1568,7 +1568,7 @@ async def default_proxy_sampling_handler(
messages: list[mcp_types.SamplingMessage],
params: mcp_types.CreateMessageRequestParams,
context: ServerRequestContext[Any, Any],
-) -> mcp_types.CreateMessageResult:
+) -> mcp_types.CreateMessageResult | mcp_types.CreateMessageResultWithTools:
"""Forward sampling request from remote server to proxy's connected clients.
Relays through the SDK session for the same reason as
@@ -1578,25 +1578,19 @@ async def default_proxy_sampling_handler(
"""
ctx = get_context()
# Deprecated upstream in SDK v2; the handshake-era relay is the one caller.
- result = await ctx.session.create_message( # ty: ignore[deprecated]
+ # Relay the request and the client's answer unchanged, so the backend sees
+ # the same model, stop reason and content as with a direct connection.
+ return await ctx.session.create_message( # ty: ignore[deprecated]
messages=list(messages),
system_prompt=params.system_prompt,
+ include_context=params.include_context,
temperature=params.temperature,
max_tokens=params.max_tokens,
stop_sequences=params.stop_sequences,
+ metadata=params.metadata,
model_preferences=params.model_preferences,
related_request_id=ctx.origin_request_id,
)
- text = (
- result.content.text if isinstance(result.content, mcp_types.TextContent) else ""
- )
- content = mcp_types.TextContent(type="text", text=text)
- return mcp_types.CreateMessageResult(
- role="assistant",
- model="fastmcp-client",
- # TODO(ty): remove when ty supports isinstance exclusion narrowing
- content=content,
- )
async def default_proxy_elicitation_handler(
diff --git a/tests/server/providers/proxy/test_proxy_client.py b/tests/server/providers/proxy/test_proxy_client.py
index 4efc7aed9..c8a6eee45 100644
--- a/tests/server/providers/proxy/test_proxy_client.py
+++ b/tests/server/providers/proxy/test_proxy_client.py
@@ -3,8 +3,10 @@ from typing import Any, cast
import pytest
from anyio import create_task_group
+from mcp import CreateMessageResult
from mcp_types import (
ElicitRequestFormParams,
+ ImageContent,
LoggingLevel,
ModelHint,
ModelPreferences,
@@ -79,15 +81,29 @@ async def _backend_sample(context: Context) -> str:
)
],
system_prompt="You love FastMCP",
+ include_context="thisServer",
temperature=0.5,
max_tokens=100,
stop_sequences=["STOP"],
+ metadata={"trace": "abc"},
model_preferences=ModelPreferences(hints=[ModelHint(name="gpt-4o")]),
related_request_id=context.origin_request_id,
)
return result.content.text if isinstance(result.content, TextContent) else ""
+async def _backend_sample_result(context: Context) -> dict[str, Any]:
+ """Issue a handshake-era `sampling/createMessage` and return the whole result."""
+ result = await context.session.create_message( # ty: ignore[deprecated]
+ messages=[
+ SamplingMessage(role="user", content=TextContent(type="text", text="Hi"))
+ ],
+ max_tokens=100,
+ related_request_id=context.origin_request_id,
+ )
+ return result.model_dump(mode="json", by_alias=True, exclude_none=True)
+
+
@pytest.fixture
def fastmcp_server():
mcp = FastMCP("TestServer")
@@ -106,6 +122,10 @@ def fastmcp_server():
) -> str:
return await _backend_sample(context)
+ @mcp.tool
+ async def sampling_result(context: Context) -> dict[str, Any]:
+ return await _backend_sample_result(context)
+
@dataclass
class Person:
name: str
@@ -235,9 +255,11 @@ class TestProxyClient:
)
]
assert params.system_prompt == "You love FastMCP"
+ assert params.include_context == "thisServer"
assert params.temperature == 0.5
assert params.max_tokens == 100
assert params.stop_sequences == ["STOP"]
+ assert params.metadata == {"trace": "abc"}
assert params.model_preferences == ModelPreferences(
hints=[ModelHint(name="gpt-4o")]
)
@@ -260,6 +282,35 @@ class TestProxyClient:
result = await client.call_tool("sampling", {})
assert result.data == "I love FastMCP"
+ @pytest.mark.parametrize(
+ "content",
+ [
+ TextContent(type="text", text="I love FastMCP"),
+ ImageContent(type="image", data="aGVsbG8=", mime_type="image/png"),
+ ],
+ ids=["text", "image"],
+ )
+ async def test_forward_sampling_response_unchanged(
+ self, proxy_server: FastMCP, content: TextContent | ImageContent
+ ):
+ """
+ The backend receives the client's sampling result as the client sent it:
+ its model, stop reason and content, not a text-only placeholder.
+ """
+ answer = CreateMessageResult(
+ role="assistant",
+ model="claude-sonnet-5-5",
+ content=content,
+ stop_reason="endTurn",
+ )
+ async with Client(
+ proxy_server, mode="legacy", sampling_handler=lambda *args: answer
+ ) as client:
+ result = await client.call_tool("sampling_result", {})
+ assert result.data == answer.model_dump(
+ mode="json", by_alias=True, exclude_none=True
+ )
+
async def test_elicit_request(self, proxy_server: FastMCP):
"""
Test that the proxy client correctly forwards the `elicit` request.