Log MCPServer handler exceptions once, by kind

docs/handlers/logging.md

@@ -70,6 +70,8 @@ went to standard error: the terminal, not the wire.
     don't want log lines, you want spans. Your server already emits them: the SDK traces every
     message with OpenTelemetry out of the box. See **[OpenTelemetry](../run/opentelemetry.md)**.
 
+You don't have to log your own handlers' crashes either. When a tool or resource function raises something unexpected, the SDK writes the `ERROR` record with the traceback for you, on its own `mcp.*` loggers; a failure you raised deliberately (`ToolError`, `ResourceNotFoundError`) is an `INFO` line instead. A prompt function that raises is an `ERROR` record too, whatever it raised. **[Handling errors](../servers/handling-errors.md#what-lands-in-your-log)** has the split. (In a test using `Client(mcp, raise_exceptions=True)`, a prompt failure is handed to your test as the exception rather than logged.)
+
 ## Recap
 
 * The MCP protocol's logging capability is deprecated by the 2026-07-28 spec and not replaced. Don't build on it.

docs/migration.md

@@ -1016,7 +1016,7 @@ except MCPError as e:
 
 ### Resource not found returns `-32602` and resource lookups raise typed exceptions (SEP-2164)
 
-Reading a missing resource now returns JSON-RPC error code `-32602` (invalid params) with the requested URI in `error.data` (`{"uri": ...}`), per [SEP-2164](https://github.com/modelcontextprotocol/modelcontextprotocol/pull/2164). Previously the server returned code `0` with no `data`. Clients can now reliably distinguish not-found from other errors; a template handler that raises `ResourceNotFoundError` (from `mcp.server.mcpserver.exceptions`) produces this same response.
+Reading a missing resource now returns JSON-RPC error code `-32602` (invalid params) with the requested URI in `error.data` (`{"uri": ...}`), per [SEP-2164](https://github.com/modelcontextprotocol/modelcontextprotocol/pull/2164). Previously the server returned code `0` with no `data`. Clients can now reliably distinguish not-found from other errors; a resource handler (static or template) that raises `ResourceNotFoundError` (from `mcp.server.mcpserver.exceptions`) produces this same response.
 
 The underlying lookups now raise typed exceptions instead of `ValueError`. `ResourceManager.get_resource()` raises `ResourceNotFoundError` when no resource or template matches the URI, and `ResourceTemplate.create_resource()` raises `ResourceError` when the template function fails. Neither subclasses `ValueError`, so callers catching `ValueError` should switch to `ResourceNotFoundError` / `ResourceError` (both importable from `mcp.server.mcpserver.exceptions`; `ResourceNotFoundError` subclasses `ResourceError`).
 

docs/servers/handling-errors.md

@@ -115,10 +115,29 @@ Send `get_author` a `title` that isn't a string and the SDK rejects it against t
 It means a whole class of `raise` statements you don't write: don't re-validate your own type hints.
 
 !!! info
-    Everything on this page is what a **client** sees, and the in-memory `Client` you'll write
-    tests with sees exactly the same thing. Even `raise_exceptions=True` doesn't turn a tool error
-    back into a traceback: by the time that flag could act, your exception is already the
-    `is_error=True` result. Assert on the result. **[Testing](../get-started/testing.md)** covers the pattern.
+    Everything so far is what a **client** sees, and the in-memory `Client` you'll write tests
+    with sees exactly the same thing. Even `raise_exceptions=True` doesn't hand a failing tool's
+    exception back to the caller: by the time that flag could act, your exception is already the
+    `is_error=True` result. Assert on the result; the traceback is in the server's log (next
+    section), which pytest's `caplog` captures. **[Testing](../get-started/testing.md)** covers the pattern.
+
+## What lands in your log
+
+Your server keeps its own record of these failures, and it draws one more line: between a failure you anticipated and one you didn't.
+
+`get_author` raised a plain `ValueError`. The model got the message, but the SDK can't know you *meant* that exception, so it assumes you didn't: the call is logged at `ERROR` with the full traceback. That is exactly what you want on the day the exception is a `KeyError` from three libraries down and the result text says only `'id'`.
+
+When the failure is one you planned for, say so with `ToolError`:
+
+```python title="server.py" hl_lines="2 12-13"
+--8<-- "docs_src/handling_errors/tutorial004.py"
+```
+
+The model reads precisely what it read before. The difference is on your side: a `ToolError` is logged as one `INFO` line with no traceback, so a production log at `WARNING` stays quiet until something is actually broken. Bad arguments and unknown tool names are `INFO` lines too; those are the caller's mistakes, not yours.
+
+Resources draw the same line. The `-32603` from a crashing resource handler names only the URI, so the `ERROR` record in your log is the one place the cause and its traceback exist. `ResourceNotFoundError`, including the SDK's own `Unknown resource`, is an `INFO` line. (A template parameter that fails its type annotation, `books://{id}` read with an `id` that isn't an `int`, currently counts as a crash.)
+
+Prompts aren't split yet: any failure in a prompt function, including an unknown name or a missing argument, is one `ERROR` record with its traceback, written by the transport layer that turns it into the JSON-RPC error.
 
 ## Recap
 
@@ -127,7 +146,8 @@ It means a whole class of `raise` statements you don't write: don't re-validate
 * The deciding question: *could a smarter model have avoided this?* Yes -> exception. No -> `MCPError`.
 * `ResourceNotFoundError` from a resource handler -> the protocol's `-32602`, with the URI in `data`.
 * Bad arguments are rejected against the schema before your function runs; you don't `raise` for those.
-* `from mcp import MCPError`; the error-code constants come from `mcp.types`.
+* In your log: an exception you didn't raise as `ToolError` is an `ERROR` record with its traceback; `ToolError`, bad tool arguments, unknown tool names, and `ResourceNotFoundError` are one `INFO` line each.
+* `from mcp import MCPError`; `ToolError` and `ResourceNotFoundError` come from `mcp.server.mcpserver.exceptions`; the error-code constants come from `mcp.types`.
 
 Errors handled. That is everything a server *exposes*. What every handler can read, and do back to the client while it runs, is the next section: **[Inside your handler](../handlers/index.md)**.
 

docs/servers/uri-templates.md

@@ -199,10 +199,11 @@ These checks are a heuristic pre-filter; for filesystem access,
 `safe_join` remains the containment boundary.
 
 !!! tip
-    If your handler can't fulfil the request (the file doesn't exist,
-    the id is unknown), raise an exception. The SDK turns it into an
-    error response. See **[Handling errors](handling-errors.md)** for the difference between a
-    protocol error and a tool error.
+    If your handler can't fulfil the request (the file doesn't exist, the id is unknown), raise
+    `ResourceNotFoundError` from `mcp.server.mcpserver.exceptions`. The client gets `-32602` with
+    your message and the URI, and your log gets one `INFO` line; any other exception is treated as
+    a crash (`-32603`, and an `ERROR` record with the traceback). See
+    **[Handling errors](handling-errors.md#a-resource-that-doesnt-exist)**.
 
 ## Resources on the low-level Server
 

docs/troubleshooting.md

@@ -92,6 +92,8 @@ result.structured_content  # None
 
 The fix is in your client: **check `result.is_error`**. A `try/except` around `call_tool` catches none of these, because there is nothing to catch. This is deliberate, and it is the single most useful thing on this page to internalise: the *model* chose the call, so the model gets the message and a chance to try again. **[Handling errors](servers/handling-errors.md)** is the whole story, including the `MCPError` path that *does* raise.
 
+If `<message>` alone doesn't tell you what broke, the traceback is in the **server's log**: an exception the tool didn't raise as `ToolError` is logged there at `ERROR`, as `Tool '<name>' raised an unexpected exception`.
+
 ## `TypeError: The @tool decorator was used incorrectly. Did you forget to call it? Use @tool() instead of @tool`
 
 You wrote `@mcp.tool` instead of `@mcp.tool()`. `tool()` is a decorator *factory*: without the parentheses, Python hands your function to its `name=` parameter.

docs_src/handling_errors/tutorial004.py

@@ -0,0 +1,14 @@
+from mcp.server import MCPServer
+from mcp.server.mcpserver.exceptions import ToolError
+
+mcp = MCPServer("Bookshop")
+
+CATALOG = {"Dune": "Frank Herbert", "Neuromancer": "William Gibson"}
+
+
+@mcp.tool()
+def get_author(title: str) -> str:
+    """Look up the author of a book in the catalog."""
+    if title not in CATALOG:
+        raise ToolError(f"No book titled {title!r} in the catalog.")
+    return CATALOG[title]

src/mcp/server/mcpserver/exceptions.py

@@ -6,20 +6,55 @@ class MCPServerError(Exception):
 
 
 class ResourceError(MCPServerError):
-    """Error in resource operations."""
+    """Error in resource operations.
+
+    When a resource or resource template handler raises this, its message reaches
+    the client as a `-32603` protocol error.
+    """
 
 
 class ResourceNotFoundError(ResourceError):
     """Resource does not exist.
 
-    Raise this from a resource template handler to signal that the requested instance does not exist;
+    Raise this from a resource handler to signal that the requested instance does not exist;
     clients receive `-32602` (invalid params) per
     [SEP-2164](https://github.com/modelcontextprotocol/modelcontextprotocol/pull/2164).
     """
 
 
+class UnexpectedResourceError(ResourceError):
+    """A resource read failed with something other than `ResourceError` or `MCPError`.
+
+    MCPServer raises this itself, around a crash in a resource or resource
+    template handler or a failed file read; you never raise it. `__cause__` is
+    the original exception, which the server logs with its traceback. The
+    message names only the URI, so the original text is withheld from the client.
+    """
+
+
 class ToolError(MCPServerError):
-    """Error in tool operations."""
+    """A tool failure the model should read.
+
+    Raise this from a tool (or a resolver) for a failure you anticipate: the
+    call returns `is_error=True` with the message in `content`, and the server
+    logs it at INFO without a traceback. Any other exception reaches the model
+    the same way but is treated as a crash and logged at ERROR with its traceback.
+
+    The SDK raises it too, for an unknown tool name and for arguments that fail
+    the input schema, and `UnexpectedToolError` subclasses it, so `except ToolError`
+    around `MCPServer.call_tool()` catches every tool failure, crash or not.
+    """
+
+
+class UnexpectedToolError(ToolError):
+    """A tool call failed with something other than `ToolError` or `MCPError`.
+
+    MCPServer raises this itself, around a crash in the tool (or a resolver) or a
+    return value that fails output conversion; you never raise it. `__cause__` is
+    the original exception, which the server logs with its traceback before
+    returning the usual `is_error=True` result. Catch it around
+    `MCPServer.call_tool()` to tell a crash from a deliberate `ToolError`.
+    """
 
 
 class InvalidSignature(Exception):

src/mcp/server/mcpserver/prompts/base.py

@@ -196,5 +196,5 @@ async def render(
             return messages
         except MCPError:
             raise
-        except Exception as e:
-            raise ValueError(f"Error rendering prompt {self.name}: {e}")
+        except Exception as exc:
+            raise ValueError(f"Error rendering prompt {self.name}: {exc}") from exc

src/mcp/server/mcpserver/resources/templates.py

@@ -11,18 +11,15 @@
 from mcp_types import Annotations, Icon, InputRequiredResult
 from pydantic import BaseModel, Field, validate_call
 
-from mcp.server.mcpserver.exceptions import ResourceError
+from mcp.server.mcpserver.exceptions import ResourceError, UnexpectedResourceError
 from mcp.server.mcpserver.resources.types import FunctionResource, Resource
 from mcp.server.mcpserver.utilities.context_injection import find_context_parameter, inject_context
 from mcp.server.mcpserver.utilities.func_metadata import func_metadata
-from mcp.server.mcpserver.utilities.logging import get_logger
 from mcp.shared._callable_inspection import is_async_callable
 from mcp.shared.exceptions import MCPError
 from mcp.shared.path_security import contains_path_traversal, is_absolute_path
 from mcp.shared.uri_template import UriTemplate
 
-logger = get_logger(__name__)
-
 if TYPE_CHECKING:
     from mcp.server.context import LifespanContextT, RequestT
     from mcp.server.mcpserver.context import Context
@@ -218,7 +215,9 @@ async def create_resource(
         carrying the echoed opaque state.
 
         Raises:
-            ResourceError: If creating the resource fails.
+            ResourceError: If the template function raises `ResourceError`.
+            UnexpectedResourceError: If the template function raises anything other
+                than `ResourceError` or `MCPError`; `__cause__` is the original.
         """
         try:
             # Add context to params if needed
@@ -247,5 +246,6 @@ async def create_resource(
         except (ResourceError, MCPError):
             raise
         except Exception as exc:
-            logger.exception(f"Error creating resource from template {uri}")
-            raise ResourceError(f"Error creating resource from template {uri}") from exc
+            # Name only the URI: the original text is withheld from the client, and
+            # the server logs the traceback from `__cause__`.
+            raise UnexpectedResourceError(f"Error creating resource from template {uri}") from exc

src/mcp/server/mcpserver/resources/types.py

@@ -16,6 +16,7 @@
 from mcp_types import Annotations, Icon, InputRequiredResult
 from pydantic import Field, validate_call
 
+from mcp.server.mcpserver.exceptions import ResourceError, UnexpectedResourceError
 from mcp.server.mcpserver.resources.base import Resource
 from mcp.shared._callable_inspection import is_async_callable
 from mcp.shared.exceptions import MCPError
@@ -79,7 +80,12 @@ class FunctionResource(Resource):
     fn: Callable[[], Any] = Field(exclude=True)
 
     async def read(self) -> str | bytes:
-        """Read the resource by calling the wrapped function."""
+        """Read the resource by calling the wrapped function.
+
+        Raises:
+            UnexpectedResourceError: If the function raises anything other than
+                `ResourceError` or `MCPError`; `__cause__` is the original.
+        """
         try:
             fn = self.fn
             if is_async_callable(fn):
@@ -103,10 +109,12 @@ async def read(self) -> str | bytes:
                 return result
             else:
                 return pydantic_core.to_json(result, fallback=str, indent=2).decode()
-        except MCPError:
+        except (MCPError, ResourceError):
             raise
-        except Exception as e:
-            raise ValueError(f"Error reading resource {self.uri}: {e}")
+        except Exception as exc:
+            # Name only the URI: the original text is withheld from the client, and
+            # the server logs the traceback from `__cause__`.
+            raise UnexpectedResourceError(f"Error reading resource {self.uri}") from exc
 
     @classmethod
     def from_function(
@@ -187,8 +195,8 @@ async def read(self) -> str | bytes:
             if self.encoding is None:
                 return await anyio.to_thread.run_sync(self.path.read_bytes)
             return await anyio.to_thread.run_sync(partial(self.path.read_text, encoding=self.encoding))
-        except Exception as e:
-            raise ValueError(f"Error reading file {self.path}: {e}")
+        except Exception as exc:
+            raise UnexpectedResourceError(f"Error reading resource {self.uri}") from exc
 
 
 class HttpResource(Resource):
@@ -232,14 +240,14 @@ def list_files(self) -> list[Path]:  # pragma: no cover
             if self.pattern:
                 return list(self.path.glob(self.pattern)) if not self.recursive else list(self.path.rglob(self.pattern))
             return list(self.path.glob("*")) if not self.recursive else list(self.path.rglob("*"))
-        except Exception as e:
-            raise ValueError(f"Error listing directory {self.path}: {e}")
+        except Exception as exc:
+            raise ValueError(f"Error listing directory {self.path}: {exc}") from exc
 
     async def read(self) -> str:  # Always returns JSON string  # pragma: no cover
         """Read the directory listing."""
         try:
             files = await anyio.to_thread.run_sync(self.list_files)
             file_list = [str(f.relative_to(self.path)) for f in files if f.is_file()]
             return json.dumps({"files": file_list}, indent=2)
-        except Exception as e:
-            raise ValueError(f"Error reading directory {self.path}: {e}")
+        except Exception as exc:
+            raise UnexpectedResourceError(f"Error reading resource {self.uri}") from exc