28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181 | class Tool(BaseModel):
"""Internal tool registration info."""
fn: Callable[..., Any] = Field(exclude=True)
name: str = Field(description="Name of the tool")
title: str | None = Field(None, description="Human-readable title of the tool")
description: str = Field(description="Description of what the tool does")
parameters: dict[str, Any] = Field(description="JSON schema for tool parameters")
fn_metadata: FuncMetadata = Field(
description="Metadata about the function including a pydantic model for tool arguments"
)
is_async: bool = Field(description="Whether the tool is async")
context_kwarg: str | None = Field(None, description="Name of the kwarg that should receive context")
resolved_params: dict[str, Any] = Field(
default_factory=lambda: {},
exclude=True,
description="Parameters filled by resolvers, mapped to (Resolve, wants_union)",
)
resolver_plans: dict[Hashable, Any] = Field(
default_factory=lambda: {}, exclude=True, description="Static per-resolver parameter plans"
)
annotations: ToolAnnotations | None = Field(None, description="Optional annotations for the tool")
icons: list[Icon] | None = Field(default=None, description="Optional list of icons for this tool")
meta: dict[str, Any] | None = Field(default=None, description="Optional metadata for this tool")
@cached_property
def output_schema(self) -> dict[str, Any] | None:
return self.fn_metadata.output_schema
@classmethod
def from_function(
cls,
fn: Callable[..., Any],
name: str | None = None,
title: str | None = None,
description: str | None = None,
context_kwarg: str | None = None,
annotations: ToolAnnotations | None = None,
icons: list[Icon] | None = None,
meta: dict[str, Any] | None = None,
structured_output: bool | None = None,
) -> Tool:
"""Create a Tool from a function."""
func_name = name or fn.__name__
validate_and_warn_tool_name(func_name)
if func_name == "<lambda>":
raise ValueError("You must provide a name for lambda functions")
func_doc = description or fn.__doc__ or ""
is_async = is_async_callable(fn)
if context_kwarg is None: # pragma: no branch
context_kwarg = find_context_parameter(fn)
resolved_params = find_resolved_parameters(fn)
if resolved_params and returns_input_required(fn):
raise InvalidSignature(
f"Tool {func_name!r} combines Resolve(...) parameters with an InputRequiredResult "
"return; a call has one input_required channel, so the multi-round flow is driven "
"either by resolvers or by the tool body, not both"
)
skip_names = [context_kwarg] if context_kwarg is not None else []
skip_names.extend(resolved_params)
func_arg_metadata = func_metadata(
fn,
skip_names=skip_names,
structured_output=structured_output,
)
parameters = func_arg_metadata.arg_model.model_json_schema(by_alias=True)
# Match `model_dump_one_level`'s kwarg keys (alias when present, else field name)
# so a by-name resolver param resolves to a key that exists at call time.
tool_arg_names = {field.alias or name for name, field in func_arg_metadata.arg_model.model_fields.items()}
resolver_plans = build_resolver_plans(resolved_params, tool_arg_names)
return cls(
fn=fn,
name=func_name,
title=title,
description=func_doc,
parameters=parameters,
fn_metadata=func_arg_metadata,
is_async=is_async,
context_kwarg=context_kwarg,
resolved_params=dict(resolved_params),
resolver_plans=resolver_plans,
annotations=annotations,
icons=icons,
meta=meta,
)
async def run(
self,
arguments: dict[str, Any],
context: Context[LifespanContextT, RequestT],
convert_result: bool = False,
) -> Any:
"""Run the tool with arguments.
Raises:
ToolError: If the tool function raises during execution.
"""
try:
pass_directly: dict[str, Any] = {}
if self.context_kwarg is not None:
pass_directly[self.context_kwarg] = context
# Resolvers see the same validated arguments the tool body receives:
# validate once and reuse it, so a `default_factory`/stateful validator
# can't hand a by-name resolver a different value than the body.
pre_validated: dict[str, Any] | None = None
if self.resolved_params:
pre_validated = self.fn_metadata.validate_arguments(arguments)
resolved = await resolve_arguments(self.resolved_params, self.resolver_plans, pre_validated, context)
if isinstance(resolved, InputRequiredResult):
# A resolver still needs client input (>= 2026-07-28): surface the
# batched questions instead of running the tool body this round.
return self.fn_metadata.convert_result(resolved) if convert_result else resolved
pass_directly |= resolved
result = await self.fn_metadata.call_fn_with_arg_validation(
self.fn,
self.is_async,
arguments,
pass_directly or None,
pre_validated=pre_validated,
)
# Registration rejects the annotated form of this combination; this covers
# a body that returns an InputRequiredResult without declaring it.
if self.resolved_params and isinstance(result, InputRequiredResult):
raise ToolError(
"the tool returned an InputRequiredResult but its parameters use Resolve(...); "
"a call has one input_required channel, so the multi-round flow is driven "
"either by resolvers or by the tool body, not both"
)
if convert_result:
result = self.fn_metadata.convert_result(result)
return result
except MCPError:
# `MCPError` (and subclasses such as `UrlElicitationRequiredError`)
# carries a JSON-RPC `ErrorData(code, message, data)` and means
# "respond with a protocol error" - re-raise so the kernel surfaces
# it as a top-level JSON-RPC error rather than wrapping it as a
# `CallToolResult(isError=True)` execution failure.
raise
except Exception as e:
raise ToolError(f"Error executing tool {self.name}: {e}") from e
|