Struct ToolServerHandle 

pub struct ToolServerHandle(/* private fields */);

A cheaply-cloneable handle to the shared tool server state.

All operations acquire locks directly on the underlying state. Multiple handles (e.g. across agents) can share the same state without channel-based message routing.

Implementations

impl ToolServerHandle

pub async fn add_tool( &self, tool: impl ToolDyn + 'static, ) -> Result<(), ToolServerError>

Register a new static tool. Re-registering an existing name replaces the implementation (last wins) and keeps its position.

pub async fn append_toolset( &self, toolset: ToolSet, ) -> Result<(), ToolServerError>

Merge an entire toolset into the server. Tool names from toolset are appended to the static-tool list in toolset’s registration order, so the tools become visible to the LLM via Self::get_tool_defs. Existing names are replaced (last wins) and keep their position.

pub async fn remove_tool(&self, tool_name: &str) -> Result<(), ToolServerError>

Remove a tool by name from both the toolset and the static list.

pub async fn call_tool( &self, tool_name: &str, args: &str, ) -> Result<String, ToolServerError>

Look up and execute a tool by name.

The tool handle is cloned under a brief read lock so that long-running tool executions never block writers.

pub async fn call_tool_with_extensions( &self, tool_name: &str, args: &str, extensions: &ToolCallExtensions, ) -> Result<String, ToolServerError>

Look up and execute a tool by name with per-call runtime extensions.

The extensions are threaded through to Tool::call_with_extensions, allowing tools to access caller-provided values (auth tokens, session IDs, etc.). The tool handle is cloned under a brief read lock so that long-running tool executions never block writers.

pub async fn call_tool_structured( &self, tool_name: &str, args: &str, extensions: &ToolCallExtensions, ) -> ToolExecutionResult

Look up and execute a tool by name, returning the structured ToolExecutionResult (model output + ToolOutcome

• result extensions).

The structured counterpart of call_tool_with_extensions, and the path the agent loop drives so hooks, tracing, and policies observe the structured outcome. A missing tool resolves to a NotFound outcome rather than a Result::Err. The tool handle is cloned under a brief read lock so that long-running tool executions never block writers.

pub async fn get_tool_defs( &self, prompt: Option<String>, ) -> Result<Vec<ToolDefinition>, ToolServerError>

Retrieve tool definitions, optionally using a prompt to select dynamic tools from configured vector stores.

Trait Implementations

impl Clone for ToolServerHandle

fn clone(&self) -> ToolServerHandle

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.