Host / client / server
The host is the AI application. Its MCP client exchanges protocol messages with Stride’s MCP server. The LLM proposes operations; the host manages their execution.
Client requests enter through MCP. Stride checks access, validates the operation, and returns structured information. The client does not bypass application rules.
The host contains the model and MCP client. Stride exposes the operations. Requests cross an authenticated connection; application permissions still apply.
The host is the AI application. Its MCP client exchanges protocol messages with Stride’s MCP server. The LLM proposes operations; the host manages their execution.
Tools execute operations. Resources expose context. Prompts provide reusable interaction templates. Stride’s current learner baseline exposes eight tools; additional primitives belong to the expansion plan.
MCP defines the message exchange. Stride defines which work an identity can access and which state changes are valid. A protocol capability does not grant organization or object access.
These references describe MCP, including its newer specification. They do not establish that every protocol feature is enabled in Stride or supported by every host.
Claude, ChatGPT, or a command-line host.
Proposes an operation using the task and conversation context.
Sends the request and returns its result to the host.
tools/callTool name + argumentsStructured resultOr an operation errorExposes bounded learner capabilities.
Requests use the connected Stride account and its allowed capabilities. A connection does not grant unrestricted access.
The response contains authorized task context or recorded status. A successful model response alone does not establish completion.
Practice stays outside the grading flow. An eligible attempt fixes the rubric; a graded submission adds evidence, scores, and reasons. Validation and a passing result are separate checks.
Practice stays outside the grading flow. An eligible attempt fixes the rubric; a graded submission adds evidence, scores, and reasons. Validation and a passing result are separate checks.
A client timeout does not prove failure.
The learner connector supports eligible work. Native activities and challenge submission stay in Stride.
| Effect | Operation | Contract |
|---|---|---|
| READ | get_my_statusConnection status | Read your connection state and recent assessment summary. |
| READ | list_current_workCurrent work | List your current eligible Stride work. |
| READ | get_work_detailWork details | Retrieve task context, available criteria and recorded feedback. |
| READ | find_microlearningsLesson search | Find accessible microlearnings beyond the current-work list. |
| READ | get_attempt_statusAttempt status | Read the recorded assessment state and outcome. |
| WRITE | start_attemptStart an attempt | Create or recover a private attempt and receive its frozen rubric. |
| WRITE | add_attempt_evidenceAttach evidence | Add a selected excerpt or supported reference to your attempt. |
| WRITE | submit_assessmentSubmit an assessment | Submit criterion scores, reasons and evidence for validation. |
These are the eight active learner tool contracts. The prepared interactive and product-help packages await validation and release. Later resource, contribution and role capabilities are planned.
Stride remains responsible for account scope, assessment validation, and recorded state. Each boundary has a specific effect.
Sign in with your own Stride account and review the connection’s access. Learner operations remain within that account’s authorized scope.
Stride receives selected work, criterion scores and reasons, and the outcome. The whole conversation is not included in the connector submission.
Stride validates submitted scores and evidence requirements against the frozen rubric. It does not independently prove every claim in the evidence.
Disconnect through connected-client controls. Revocation ends future access; it does not undo recorded work or retract prior responses from a client.