modelcontextprotocol / modelcontextprotocol/php-sdk
[Server] SEP-2663 Tasks: Handlers return task handles; the application advances the task
Nobody has claimed this yet.
- Dominant language
- PHP
- Stars
- 1.6k
- Forks
- 173
- Avg merge
- 2d 49m
- Merged PRs (30d)
- 23
Description
Execution model for SEP-2663 for the 2026-07-28 release. Tracked by umbrella #335.
Decision
The SDK owns storage and the tasks/* surface; advancing a task is the application's job. PHP has no in-process background executor to build on, and picking a queue or process model on the user's behalf would be the wrong call for a framework-agnostic SDK.
Scope
- Any tool, prompt or resource handler may return a
CreateTaskResult(viaTaskContext::create()) instead of its usual result; the core handlers pass anyResultInterfacea handler returns through untouched, so no core code knows about tasks. - What happens next is application code: enqueue something a worker picks up, and have the worker call
$store->save($task->with(...))as it progresses.TaskContext::getStore()gives a handler the store for a task it did not create in this request. TaskInputHandlerInterfaceis the hook for what an answer to a parked (input_required) task means;TasksUpdateHandlerdelegatesinputResponsesto it.- No
#[McpTool]task-eligibility marker and no early-return path inReferenceHandler: whether to create a task is the handler's decision, made per request.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reading TaskContext::create(), TaskContext::getStore(), TaskInputHandlerInterface, and TasksUpdateHandler. Trace how handlers return ResultInterface values and how the store saves task progress; done means task results pass through, parked-task input delegates correctly, and application code can advance tasks without core task-specific logic.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- php
- Domain
- api, backend
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100