Tasks, Queues & Durable Workflows API¶

API reference for the MCP Tasks protocol runtime (MCPTaskExecutor, MCPTaskRuntime, MCPTaskStore, TaskRecord, ToolTaskCall, TaskLookupError, TaskStateError) and the litestar_mcp.queues distributed execution layer (QueueTaskExecutor, DurableWorkflowEngine, WorkflowStore, WorkflowRun, NodeRecord, TaskDependencyChain, QueuedAgentChatController, QueuedTurnManager, QueuedTurn, AgentRegistry, PrincipalSnapshot).

For architectural guides, see MCP Tasks & Human-in-the-Loop, Distributed Task Queues, Durable DAG Workflows, and Queued Agent Turns.

MCP Tasks Store & Executor Protocol (`io.modelcontextprotocol/tasks`)¶

MCPTaskExecutor ¶

Bases: Protocol

Runs task-augmented tool calls outside the request that created them.

LitestarMCP calls configure during app init so the executor can wire itself into other plugins, and bind at startup with the collaborators that need the running application.

configure ¶

configure(app_config)

Register executor hooks or dependencies on the Litestar AppConfig during initialization.

bind ¶

bind(runtime)

Bind the running application, task store, and attempt runner callback at startup.

submit ¶

submit(record, call)

Enqueue or schedule initial execution of a newly created task-augmented tool call.

resume ¶

resume(call)

Schedule a resumed execution attempt after client input responses are received.

cancel ¶

cancel(record)

Cancel an in-flight task execution backed by this executor.

MCPTaskRuntime ¶

Collaborators an executor needs once the application has started.

Parameters:

Name	Type	Description	Default
`app`	`Litestar`	The running Litestar application instance.	required
`task_store`	`MCPTaskStore`	The active `MCPTaskStore` managing persisted task records.	required
`run_attempt`	`Callable[list, Awaitable[None]]`	Async callback that executes a single `ToolTaskCall` attempt against the handler service.	required

Attributes:

Name	Type	Description
`app`	`Litestar`	The running Litestar application instance.
`task_store`	`MCPTaskStore`	The active `MCPTaskStore` managing persisted task records.
`run_attempt`	`Callable[[ToolTaskCall, Any], Awaitable[None]]`	Async callback that executes a single `ToolTaskCall` attempt against the handler service.

MCPTaskStore ¶

Task records backed by a Litestar Store.

Only serializable records are persisted. Runner tasks, cancellation flags, and input queues are process-local coordination primitives; production deployments use a shared Store for retrieval and a shared notification backend for fan-out.

set_status_callback ¶

set_status_callback(callback)

Set the task-state notification callback.

create ¶

create(owner_id, ttl_ms=None, *, local=True)

Durably create a task before returning its handle.

attach_runner ¶

attach_runner(task_id, runner)

Attach a process-local runner to a durable task.

has_local_runner ¶

has_local_runner(task_id)

Return whether a process-local runner is attached to task_id.

set_job_id ¶

set_job_id(task_id, job_id)

Persist the executor's job id for the task's current attempt under a separate key.

get_job_id ¶

get_job_id(task_id)

Retrieve the executor's job id for the task's current attempt.

set_dispatch ¶

set_dispatch(task_id, call)

Save the serializable dispatch envelope on the record before executor submission.

get_dispatch ¶

get_dispatch(task_id)

Retrieve and decode the serializable dispatch envelope.

get ¶

get(task_id, owner_id)

Retrieve a task, enforcing authenticated ownership when present.

complete ¶

complete(task_id, result)

Persist a successful tool result.

fail ¶

fail(task_id, error, status_message=None)

Persist a failed task.

require_input ¶

require_input(task_id, input_requests, request_state)

Pause a task until the client supplies input via tasks/update.

wait_for_input ¶

wait_for_input(task_id)

Wait for the next input response or cancellation request.

update ¶

update(task_id, owner_id, input_responses)

Submit client responses to a task waiting for input.

cancel ¶

cancel(task_id, owner_id)

Request cooperative cancellation and acknowledge immediately.

mark_cancelled ¶

mark_cancelled(task_id)

Persist cancellation after the runner cooperates.

close ¶

close()

Cancel all process-local runners during graceful shutdown.

TaskRecord ¶

Persisted Tasks extension state.

Parameters:

Name	Type	Description	Default
`task_id`	`str`	Unique task identifier (`taskId`).	required
`owner_id`	`str \| None`	Optional owner identifier restricting task access to the creating principal.	required
`status`	`str`	Current MCP task lifecycle status (`working`, `input_required`, `completed`, `failed`, or `cancelled`).	required
`created_at`	`datetime`	UTC timestamp when the task record was created.	required
`last_updated_at`	`datetime`	UTC timestamp of the most recent task state transition.	required
`ttl_ms`	`int \| None`	Time-to-live in milliseconds for the persisted task record.	required
`poll_interval_ms`	`int \| None`	Suggested client polling interval in milliseconds.	required
`status_message`	`str \| None`	Optional human-readable status message describing current task progress.	`None`
`input_requests`	`dict[str, dict[str, Any]] \| None`	Pending multi-round-trip input request descriptors when `status == "input_required"`.	`None`
`request_state`	`str \| None`	Opaque round-tripped state token preserved across `input_required` pauses.	`None`
`result`	`dict[str, Any] \| None`	Final MCP tool result dictionary when `status == "completed"`.	`None`
`error`	`JSONRPCError \| None`	Terminal JSON-RPC error object when `status == "failed"`.	`None`
`dispatch`	`dict[str, Any] \| None`	Serialized `ToolTaskCall` payload persisted for external executor resume.	`None`
`input_responses`	`dict[str, Any] \| None`	Most recent client input responses submitted via `tasks/update`.	`None`

Attributes:

Name	Type	Description
`task_id`	`str`	Unique task identifier (`taskId`).
`owner_id`	`str \| None`	Optional owner identifier restricting task access to the creating principal.
`status`	`str`	Current MCP task lifecycle status (`working`, `input_required`, `completed`, `failed`, or `cancelled`).
`created_at`	`datetime`	UTC timestamp when the task record was created.
`last_updated_at`	`datetime`	UTC timestamp of the most recent task state transition.
`ttl_ms`	`int \| None`	Time-to-live in milliseconds for the persisted task record.
`poll_interval_ms`	`int \| None`	Suggested client polling interval in milliseconds.
`status_message`	`str \| None`	Optional human-readable status message describing current task progress.
`input_requests`	`dict[str, dict[str, Any]] \| None`	Pending multi-round-trip input request descriptors when `status == "input_required"`.
`request_state`	`str \| None`	Opaque round-tripped state token preserved across `input_required` pauses.
`result`	`dict[str, Any] \| None`	Final MCP tool result dictionary when `status == "completed"`.
`error`	`JSONRPCError \| None`	Terminal JSON-RPC error object when `status == "failed"`.
`dispatch`	`dict[str, Any] \| None`	Serialized `ToolTaskCall` payload persisted for external executor resume.
`input_responses`	`dict[str, Any] \| None`	Most recent client input responses submitted via `tasks/update`.

is_terminal ¶

is_terminal()

Return whether the task has reached a terminal state.

to_dict ¶

to_dict()

Return the extension wire representation.

ToolTaskCall ¶

Bases: Struct

Serializable dispatch envelope for one task-augmented tools/call attempt.

Parameters:

Name	Type	Description	Default
`task_id`	`str`	Unique identifier of the task record in `MCPTaskStore`.	required
`tool_name`	`str`	Registered name of the MCP tool being executed.	required
`arguments`	`dict[str, Any]`	Tool invocation arguments supplied by the caller.	required
`owner_id`	`str \| None`	Optional task-owner identifier scoping access to the task record.	`None`
`principal`	`dict[str, Any] \| None`	Serialized principal snapshot captured from the creating request.	`None`
`queue`	`str \| None`	Optional target `litestar-queues` queue name override.	`None`
`client_id`	`str`	Identifier of the originating MCP client.	`'queue'`
`protocol_version`	`str`	Negotiated MCP protocol version for the task execution.	`'2026-07-28'`
`client_capabilities`	`dict[str, Any] \| None`	Client capabilities dictionary captured from the creating request.	`None`
`client_info`	`dict[str, Any] \| None`	Client implementation metadata captured from the creating request.	`None`
`input_responses`	`dict[str, Any] \| None`	Accumulated interactive input responses supplied via `tasks/update`.	`None`
`request_state`	`str \| None`	Opaque round-tripped state token accompanying `input_responses`.	`None`
`attempt`	`int`	1-based execution attempt counter incremented on each `tasks/update` resume.	`1`

Attributes:

Name	Type	Description
`task_id`	`str`	Unique identifier of the task record in `MCPTaskStore`.
`tool_name`	`str`	Registered name of the MCP tool being executed.
`arguments`	`dict[str, Any]`	Tool invocation arguments supplied by the caller.
`owner_id`	`str \| None`	Optional task-owner identifier scoping access to the task record.
`principal`	`dict[str, Any] \| None`	Serialized principal snapshot captured from the creating request.
`queue`	`str \| None`	Optional target `litestar-queues` queue name override.
`client_id`	`str`	Identifier of the originating MCP client.
`protocol_version`	`str`	Negotiated MCP protocol version for the task execution.
`client_capabilities`	`dict[str, Any] \| None`	Client capabilities dictionary captured from the creating request.
`client_info`	`dict[str, Any] \| None`	Client implementation metadata captured from the creating request.
`input_responses`	`dict[str, Any] \| None`	Accumulated interactive input responses supplied via `tasks/update`.
`request_state`	`str \| None`	Opaque round-tripped state token accompanying `input_responses`.
`attempt`	`int`	1-based execution attempt counter incremented on each `tasks/update` resume.

TaskLookupError ¶

Bases: ValueError

Raised when a task cannot be found or accessed.

TaskStateError ¶

Bases: ValueError

Raised when a task transition is invalid.

Distributed Queue Task Execution (`litestar_mcp.queues`)¶

QueueTaskExecutor ¶

Run MCP Tasks attempts on litestar-queues.

LitestarMCP calls configure during app init, which attaches the executor's task dependencies to the application's QueuePlugin; no queue wiring is needed beyond registering QueuePlugin on the same app.

Parameters:

Name	Type	Description	Default
`principal_loader`	`Callable[[str], Awaitable[Any]] \| None`	Optional callback rehydrating `scope["user"]` from a task `owner_id` in the worker. When omitted, `LitestarMCP` automatically rehydrates `scope["user"]`, `scope["auth"]`, and `scope["litestar_security"]` from the authenticated principal snapshot and the application's existing Litestar dependencies.	`None`
`queue`	`str`	Default queue name for tool attempts (overridable per controller or route via `mcp_queue` / `@mcp_tool(queue=...)`).	`'mcp'`
`timeout`	`int`	Per-attempt timeout in seconds.	`3600`

Attributes:

Name	Type	Description
`runtime`	`MCPTaskRuntime`	Collaborators bound at application startup.

configure ¶

configure(app_config)

Attach the executor's task dependencies to the application's QueuePlugin.

bind ¶

bind(runtime)

Bind runtime collaborators to the executor.

resolve_task_dependencies ¶

resolve_task_dependencies(task, record=None, task_context=None)

Inject this executor into tool task attempts.

Returns:

Type	Description
`Mapping[str, object]`	`{"mcp_task_executor": self}` for tool attempts, otherwise an empty mapping.

submit ¶

submit(record, call)

Enqueue the first tool call attempt on litestar-queues.

resume ¶

resume(call)

Enqueue the next attempt after client input responses arrive.

cancel ¶

cancel(record)

Cancel the task's current attempt and persist cancelled when this replica wins.

load_principal ¶

load_principal(owner_id)

Rehydrate user principal from owner_id.

run_mcp_tool_task ¶

run_mcp_tool_task(call, mcp_task_executor)

Run one MCP Tasks attempt in the worker and return the record status.

Durable DAG Workflow Engine & Persistence¶

DurableWorkflowEngine ¶

Bases: InitPluginProtocol

Runs DynamicWorkflow definitions on litestar-queues with Store-backed state.

Register the engine as a Litestar plugin next to QueuePlugin. On app init it attaches its task dependencies to the queue configuration, resolves its store from the application's store registry, and exposes itself to route handlers as the workflow_engine dependency.

Attributes:

Name	Type	Description
`store`	`WorkflowStore`	Durable run and node records.

on_app_init ¶

on_app_init(app_config)

Wire task dependencies, the default store, the workflow_engine dependency, and app capture.

Returns:

Type	Description
`AppConfig`	The updated application configuration.

resolve_task_dependencies ¶

resolve_task_dependencies(task, record=None, task_context=None)

Inject this engine into the workflow tasks.

Returns:

Type	Description
`Mapping[str, object]`	`{"workflow_engine": self}` for workflow tasks, otherwise an empty mapping.

start ¶

start(workflow, initial_state=None)

Start a new durable workflow run and enqueue initial advance task.

Parameters:

Name	Type	Description	Default
`workflow`	`str`	Registered workflow name.	required
`initial_state`	`dict[str, Any] \| None`	Optional initial state dictionary.	`None`

Returns:

Type	Description
`str`	The generated run_id string.

Raises:

Type	Description
`ImproperlyConfiguredException`	If immediate execution backend is used with watchdog.

snapshot ¶

snapshot(run_id)

Load the durable run record and reconstructed WorkflowContext.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required

Returns:

Type	Description
`tuple[WorkflowRun, WorkflowContext]`	Tuple of (WorkflowRun, WorkflowContext).

Raises:

Type	Description
`KeyError`	If the workflow run does not exist.

advance ¶

advance(run_id)

Advance a durable workflow run by scheduling ready nodes or finalizing terminal state.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required

Returns:

Type	Description
`RunStatus \| None`	The updated RunStatus, or None if run does not exist.

execute_node ¶

execute_node(run_id, node)

Execute a single workflow node, record its outcome, and re-enqueue advance.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required
`node`	`str`	Target node name to execute.	required

watch ¶

watch(run_id, tick=1)

Execute one watchdog tick for a durable run.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required
`tick`	`int`	Current tick iteration number.	`1`

Returns:

Type	Description
`RunStatus \| None`	The resulting RunStatus, or None if run does not exist.

WorkflowStore ¶

Store-backed run and node records with single-writer keys.

create_run ¶

create_run(run)

Create a new durable workflow run record.

Parameters:

Name	Type	Description	Default
`run`	`WorkflowRun`	WorkflowRun record to persist.	required

Raises:

Type	Description
`ValueError`	If a run with the same run_id already exists.

get_run ¶

get_run(run_id)

Retrieve a workflow run record by run_id.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required

Returns:

Type	Description
`WorkflowRun \| None`	The decoded WorkflowRun record or None if not found.

set_run_status ¶

set_run_status(run_id, status, *, error=None)

Update the status of a workflow run.

Terminal statuses ('completed' and 'failed') are monotonic and will not be overwritten by subsequent calls.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required
`status`	`RunStatus`	New RunStatus for the run.	required
`error`	`str \| None`	Optional error message when transitioning to failed.	`None`

Returns:

Type	Description
`WorkflowRun`	The updated (or unchanged terminal) WorkflowRun record.

Raises:

Type	Description
`KeyError`	If the workflow run does not exist.

put_node ¶

put_node(run_id, record)

Persist a write-once node record for a workflow run.

Encodes output first, propagating serialization errors. Existing node records are not overwritten.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required
`record`	`NodeRecord`	NodeRecord to store.	required

get_node ¶

get_node(run_id, node)

Retrieve a node record by run_id and node name.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required
`node`	`str`	Node name.	required

Returns:

Type	Description
`NodeRecord \| None`	The decoded NodeRecord or None if not found.

get_nodes ¶

get_nodes(run_id, names)

Retrieve multiple node records by names.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique workflow run identifier.	required
`names`	`Iterable[str]`	Iterable of node names to fetch.	required

Returns:

Type	Description
`dict[str, NodeRecord]`	Dictionary mapping node names to their decoded NodeRecord instances.

to_context ¶

to_context(run, nodes)

Reconstruct an in-process WorkflowContext from durable run and node records.

Parameters:

Name	Type	Description	Default
`run`	`WorkflowRun`	WorkflowRun record providing run_id and initial state.	required
`nodes`	`Mapping[str, NodeRecord]`	Mapping of node names to NodeRecord outcomes.	required

Returns:

Type	Description
`WorkflowContext`	Reconstructed WorkflowContext instance.

WorkflowRun ¶

Bases: Struct

Durable state of one workflow run.

Parameters:

Name	Type	Description	Default
`run_id`	`str`	Unique identifier of the workflow run.	required
`workflow`	`str`	Name of the `DynamicWorkflow` definition being executed.	required
`state`	`dict[str, Any]`	dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)	`<class 'dict'>`
`status`	`Literal['running', 'completed', 'failed']`	Current lifecycle status of the workflow run (`running`, `completed`, or `failed`).	`'running'`
`error`	`str \| None`	Optional error description recorded when `status` transitions to `failed`.	`None`
`created_at`	`float`	time() -> floating-point number Return the current time in seconds since the Epoch. Fractions of a second may be present if the system clock provides them.	`1793739870.647042`

Attributes:

Name	Type	Description
`run_id`	`str`	Unique identifier of the workflow run.
`workflow`	`str`	Name of the `DynamicWorkflow` definition being executed.
`state`	`dict[str, Any]`	Initial workflow state persisted when the run was created.
`status`	`RunStatus`	Current lifecycle status of the workflow run (`running`, `completed`, or `failed`).
`error`	`str \| None`	Optional error description recorded when `status` transitions to `failed`.
`created_at`	`float`	Unix epoch timestamp (seconds) when the workflow run was created.

NodeRecord ¶

Bases: Struct

Write-once outcome of one node in one run.

Parameters:

Name	Type	Description	Default
`node`	`str`	Name of the workflow node within the DAG.	required
`status`	`Literal['completed', 'skipped', 'failed']`	Terminal step status (`completed`, `skipped`, or `failed`).	required
`output`	`Any`	JSON-serializable output produced by the node when completed.	`None`
`error`	`str \| None`	Optional error description recorded when the node failed.	`None`

Attributes:

Name	Type	Description
`node`	`str`	Name of the workflow node within the DAG.
`status`	`StepStatus`	Terminal step status (`completed`, `skipped`, or `failed`).
`output`	`Any`	JSON-serializable output produced by the node when completed.
`error`	`str \| None`	Optional error description recorded when the node failed.

TaskDependencyChain ¶

Ordered task dependency resolvers merged into one resolver.

Later resolvers win on key collisions. QueueConfig accepts a single task_dependency_resolver, so integrations append themselves to the application's queue configuration through attach instead of asking users to combine resolvers by hand.

attach ¶

attach(app_config, resolver, *, owner)

Add resolver to the QueuePlugin registered on app_config.

The queue reads task_dependency_resolver on every attempt, so the order in which plugins initialize does not matter. Attaching the same resolver twice is a no-op. An existing user resolver is kept and runs before resolver.

Parameters:

Name	Type	Description	Default
`app_config`	`AppConfig`	Application configuration being initialized.	required
`resolver`	`TaskDependencyResolver`	Resolver supplying the integration's task keyword arguments.	required
`owner`	`str`	Integration name used in configuration error messages.	required

Returns:

Type	Description
`QueuePlugin`	The application's `QueuePlugin`.

Raises:

Type	Description
`ImproperlyConfiguredException`	If no `QueuePlugin` is registered, or the queue uses a `task_dependency_provider`, which litestar-queues does not allow alongside a resolver.

Queued Agent Turns & Streaming Bridge¶

QueuedAgentChatController ¶

Bases: Controller

Litestar controller enqueuing agent turns and streaming frames from Channels.

Attributes:

Name	Type	Description
`replay_limit`	`int`	Default: `1000`
`ping_interval`	`float`	Default: `15.0`

enqueue_turn ¶

enqueue_turn(data, request, turn_manager)

Enqueue a background agent turn and return HTTP 202 Accepted with stream URL.

stream_turn ¶

stream_turn(turn_id, request, channels=None)

Stream real-time SSE frames for a queued agent turn from Litestar Channels.

emit_agent_frame ¶

emit_agent_frame(channels, turn_id, frame, *, seq=None, type_encoders=None)

Serialize and publish an AgentStreamFrame to the turn's Litestar Channel.

agent_turn_channel ¶

agent_turn_channel(turn_id)

Return the canonical Litestar Channels topic for a queued agent turn.

QueuedTurnManager ¶

Manages queuing agent turns and publishing realtime frames to Channels.

enqueue_turn ¶

enqueue_turn(
    session_id,
    user_message,
    *,
    turn_id=None,
    agent_name="root",
    dynamic_context=None,
    principal=None,
)

Enqueue an agent turn for background execution.

QueuedTurn ¶

Bases: Struct

Serializable envelope for background turn execution.

Parameters:

Name	Type	Description	Default
`session_id`	`str`	Conversation session identifier for the queued turn.	required
`turn_id`	`str`	Unique identifier assigned to the queued turn.	required
`user_message`	`str \| dict[str, Any]`	User message string or structured payload initiating the turn.	required
`agent_name`	`str`	Registered agent runtime name in `AgentRegistry` that will execute the turn.	`'root'`
`dynamic_context`	`str \| None`	Optional per-turn context prepended to the agent's system instructions.	`None`
`principal`	`PrincipalSnapshot \| None`	Optional serialized security principal snapshot for worker rehydration.	`None`

Attributes:

Name	Type	Description
`session_id`	`str`	Conversation session identifier for the queued turn.
`turn_id`	`str`	Unique identifier assigned to the queued turn.
`user_message`	`str \| dict[str, Any]`	User message string or structured payload initiating the turn.
`agent_name`	`str`	Registered agent runtime name in `AgentRegistry` that will execute the turn.
`dynamic_context`	`str \| None`	Optional per-turn context prepended to the agent's system instructions.
`principal`	`PrincipalSnapshot \| None`	Optional serialized security principal snapshot for worker rehydration.

AgentRegistry ¶

Registry mapping agent names to AgentRuntime instances for worker tasks.

register ¶

register(name, runtime)

Register an AgentRuntime under an agent name.

get ¶

get(name)

Get an AgentRuntime by agent name.

PrincipalSnapshot ¶

Bases: Struct

Snapshot of security principal attributes for queued execution.

Parameters:

Name	Type	Description	Default
`user_id`	`str \| None`	Authenticated user identifier captured at enqueue time.	`None`
`tenant`	`str \| None`	Optional tenant identifier associated with the principal.	`None`
`roles`	`list[str]`	Built-in mutable sequence. If no argument is given, the constructor creates a new empty list. The argument must be an iterable if specified.	`<dynamic>`
`scopes`	`list[str]`	Built-in mutable sequence. If no argument is given, the constructor creates a new empty list. The argument must be an iterable if specified.	`<dynamic>`

Attributes:

Name	Type	Description
`user_id`	`str \| None`	Authenticated user identifier captured at enqueue time.
`tenant`	`str \| None`	Optional tenant identifier associated with the principal.
`roles`	`list[str]`	Role names granted to the principal.
`scopes`	`list[str]`	Authorization scopes granted to the principal.

run_agent_turn ¶

run_agent_turn = task(
    "litestar_mcp.agent.run_turn", queue="agent", retries=0, timeout=1800
)(_run_agent_turn_body)

Tasks, Queues & Durable Workflows API¶

MCP Tasks Store & Executor Protocol (io.modelcontextprotocol/tasks)¶

MCPTaskExecutor ¶

configure ¶

bind ¶

submit ¶

resume ¶

cancel ¶

MCPTaskRuntime ¶

MCPTaskStore ¶

set_status_callback ¶

create ¶

attach_runner ¶

has_local_runner ¶

set_job_id ¶

get_job_id ¶

set_dispatch ¶

get_dispatch ¶

get ¶

complete ¶

fail ¶

require_input ¶

wait_for_input ¶

update ¶

cancel ¶

mark_cancelled ¶

close ¶

TaskRecord ¶

is_terminal ¶

to_dict ¶

ToolTaskCall ¶

TaskLookupError ¶

TaskStateError ¶

Distributed Queue Task Execution (litestar_mcp.queues)¶

QueueTaskExecutor ¶

configure ¶

bind ¶

resolve_task_dependencies ¶

submit ¶

resume ¶

cancel ¶

load_principal ¶

run_mcp_tool_task ¶

Durable DAG Workflow Engine & Persistence¶

DurableWorkflowEngine ¶

on_app_init ¶

resolve_task_dependencies ¶

start ¶

snapshot ¶

advance ¶

execute_node ¶

watch ¶

WorkflowStore ¶

create_run ¶

get_run ¶

set_run_status ¶

put_node ¶

get_node ¶

get_nodes ¶

to_context ¶

WorkflowRun ¶

NodeRecord ¶

TaskDependencyChain ¶

attach ¶

Queued Agent Turns & Streaming Bridge¶

QueuedAgentChatController ¶

enqueue_turn ¶

stream_turn ¶

emit_agent_frame ¶

agent_turn_channel ¶

QueuedTurnManager ¶

enqueue_turn ¶

QueuedTurn ¶

AgentRegistry ¶

register ¶

get ¶

PrincipalSnapshot ¶

run_agent_turn ¶

MCP Tasks Store & Executor Protocol (`io.modelcontextprotocol/tasks`)¶

Distributed Queue Task Execution (`litestar_mcp.queues`)¶