AI Connectors

What the connector is

Stride's Model Context Protocol (MCP) connector gives Claude Code or Codex access to your own Stride status and work. It requests six permissions: stride.profile.read, stride.work.read, stride.attempt.read, stride.attempt.write, stride.evidence.write, and stride.assessment.submit. None of them lets it see anyone else's data, publish anything, mark your work complete, or pay a reward.

The connector only ever sees the data of the person who connected it. It cannot see a peer's, a direct report's, or anyone else's Stride status.

Connect from Claude

Hosted Claude (Claude.ai) is planned and not yet enabled, so there is no Claude connector to set up yet. When it is enabled, use the Claude connector your organization has approved. An administrator enabling Claude in Stride does not automatically provision the connector in your Claude workspace.

• Use your organization's managed connector when its owner has made Stride available.
• Connect your own Stride account and review consent. Do not assume a public directory listing is available for your organization.

Connect from Claude Code and Codex

Claude Code and Codex use Stride's remote MCP server with the client registration and authentication method approved for your organization. Copy the environment-specific server URL from your setup quest rather than using an address from another environment.

The published pilot server address is https://stride-mcp-pilot.vercel.app/api/mcp. Use the server URL shown in your setup quest for your organization; a different environment can use a different address.

Stride uses pre-registered CLI client IDs and refuses dynamic client registration. The pilot registrations are anthropic-claude-code-pilot-01 for Claude Code and openai-codex-pilot-01 for Codex. Use the approved client ID and server URL for the same environment; your admin supplies the DEV configuration.

If your administrator announces a server-address change, manually configured CLI clients can reconnect once using the new approved details. For Anthropic's Connectors Directory, an origin change requires a new listing; an existing listing does not silently move to another domain.

1. Ask your administrator for the approved tool configuration and client registration.
2. Add the remote Stride server using that configuration.
3. Sign in to your own Stride account, review consent, and run the setup quest's connection test.

What you get from each tool

Once connected, the assistant can call seven Stride tools on your behalf. The first four only read your own status and work. The next three can start, add evidence to, and submit a provisional score for your own assessment attempt. None of the seven can publish anything, mark your work complete, or pay a reward.

• get_my_status: a quick snapshot of your connection, plus how many of your assessment attempts are currently open.
• list_current_work: the weekly challenges and quest steps you're currently eligible to work on in Stride.
• get_work_detail: one work item in full, including its linked micro-learning when available, looked up by id for something you're currently eligible for.
• get_attempt_status: the current state of one specific attempt you started through the connector: open, in review, or settled.
• start_attempt: starts (or safely resumes) one private assessment attempt for an eligible work item on your behalf, and freezes its approved rubric.
• add_attempt_evidence: attaches your own evidence to one of your open attempts. Evidence stays unverified until Stride's own policy checks complete.
• submit_assessment: submits criterion-level scores and evidence as a provisional assessment for one of your open attempts, or safely defers it to a Stride human reviewer. This never creates a verified completion or reward by itself.

What admins and executives see

Each learner connects their own account. Personal connection details and assistant tool outputs are not added to ordinary manager, executive, or admin dashboards. Completing a setup lesson records normal learning progress under your organization's existing privacy settings. You can see and revoke your own connected clients.

There is nothing in flight from assistant tool calls for ordinary manager, executive, or admin dashboards to see. Setup lesson completion still records normal learning progress under your organization's existing privacy settings.

Disconnect your own access

You can revoke Stride's connector access yourself at any time, regardless of which tool you connected it to.

• Open Connected AI clients using the link below and sign in with the Stride account you connected, then select Disconnect next to the client you want to remove. This revokes access immediately, no matter which tool created it.
• In Claude Code, you can also run claude mcp logout stride to clear your local credentials.
• Once hosted Claude (Claude.ai) is enabled, you can remove the connector itself from Customize → Connectors in Claude.

How admins revoke access

There is no separate switch just for the MCP connector. A Client Admin revokes it the same way they manage any other access: deactivating a person, or otherwise changing their organization membership, role, or invitation status in Organization and users. The moment that status changes, Stride automatically revokes every MCP connector grant that person holds.

Tokens and privacy

On this deployment, access tokens last 15 minutes and refresh tokens rotate with a 30-day lifetime. The connecting tool uses this to renew your access automatically, without asking you to sign in again.

For how Stride collects, stores, and protects this connector's data, see Stride's MCP connector privacy statement.

Questions about this connector? Contact ryan.cappelluti@cgi.com.

Enterprise managed auth (admin setup)

First enable the intended providers and specific tools in Stride Admin Settings. Then provision the connector in the external workspace and confirm the approved registration and authenticated connection before marking that client available for setup.

Learners still connect their own Stride accounts and approve consent. A saved provider selection alone is not proof that a workspace connector works.

• In your Admin sidebar, Setup (AI connectors) shows the server URL for your environment, vendor setup notes for the providers you turned on (not every one of those assistants can connect to Stride yet), and downloadable skill packages.

1. Choose the providers and sub-tools your organization uses in AI Providers and Tools.
2. Follow the current vendor guide to approve the Stride connector for the intended workspace members.
3. Use the reviewed OAuth client registration and the server URL for the target environment.
4. Test sign-in and both read tools using a real member account. Confirm network access and workspace restrictions.
5. Ask your Stride support contact to confirm the approved connection test before making the connector available to learners.

Your first setup quest

After goal selection, choose Connect my assistant to start the setup quest, or Connect later to continue your first AI win. The setup quest remains pinned on your dashboard.

If several assistants are enabled, choose one to start. A verified connection unlocks eligible connected work; completing all three lessons finishes the setup quest. These are separate states.

1. Learn what the connection does and which account it can access.
2. Connect one enabled assistant and run the connection check.
3. Practice finding Stride work and understanding provisional assessment and verified completion.

Set up ChatGPT (not available yet)

No Stride client for ChatGPT exists today, so there are no ChatGPT setup steps. Stride's connector works with Claude Code and Codex, which are in live use on pilot. Hosted Claude (Claude.ai) is planned and not yet enabled.

Set up Codex

Use your organization's approved Codex installation and remote MCP configuration. Client registration may differ between organizations; use the details your administrator supplies.

For the published pilot environment, use the pre-registered Codex client below. Follow the command as shown without adding --oauth-resource to this Stride configuration. Use your admin's environment-specific registration for DEV.

1. Add Stride using the server URL shown in your setup quest and the approved configuration.
2. Sign in with your own Stride account through the MCP authentication flow and review consent.
3. Return to Codex and run the supplied connection test.
4. codex mcp add stride --url https://stride-mcp-pilot.vercel.app/api/mcp --oauth-client-id openai-codex-pilot-01
5. codex mcp login stride --scopes mcp.connect
6. To disconnect this CLI connection, run codex mcp logout stride and revoke its grant in Connected AI clients.

Set up hosted Claude (not yet available)

Hosted Claude (Claude.ai) is planned and not yet enabled. Its client is registered, but Stride has not yet enabled or tested that connection surface end to end, so there are no Claude setup steps yet.

For now, set up Claude Code or Codex instead. See Set up Claude Code and Connect from Claude Code and Codex in this article.

Set up Claude Code

Use the approved Claude Code installation and the remote-server configuration supplied by your organization.

For the published pilot environment, use the pre-registered Claude Code client below. Use your admin's environment-specific registration for DEV.

1. Add the Stride MCP server using the URL shown in your setup quest and the approved client registration.
2. Use the tool's MCP sign-in flow to connect your own Stride account and review consent.
3. Return to Claude Code and run the supplied connection test.
4. claude mcp add --transport http --client-id anthropic-claude-code-pilot-01 stride https://stride-mcp-pilot.vercel.app/api/mcp
5. claude mcp login stride
6. To disconnect this CLI connection, run claude mcp logout stride and revoke its grant in Connected AI clients.

Test your connection

Choose Start connection check in the setup quest, then copy the generated test prompt into your selected assistant. It asks Stride to read your status and list your current work using the same setup check ID.

Return to Stride and choose Check result. Connection verified means both calls succeeded through the same approved client and active account grant. An empty work list is a successful result. The check lasts 15 minutes; start another if it expires.

Signing in, opening a guide, an assistant claiming success, or a last-used timestamp does not by itself verify the connection. The connection test does not start an assessment, complete learning work, publish content, or award rewards.

If setup does not work

• Stride is missing: confirm your workspace and ask its admin to enable the approved connector.
• Sign-in fails: use your own Stride account in the correct organization, then retry consent.
• Trouble signing in: sign in to the main Stride site first. If that fails, select Forgot password there, reset your password, and then add the connector again. If the main site works but the connector still does not, reset your password on the main site anyway. Forgot password on the connector's own sign-in page does not work.
• No calls are recorded: select or enable Stride in the assistant conversation and send the generated test prompt.
• The check expired: start a new check and use its new prompt.
• Wrong assistant or account: return to Stride, select the intended enabled assistant, and start a new check.
• Reconnect required: reauthorize and verify your connection again. Previously completed lessons remain completed.
• Setup service unavailable: retry later. Ordinary in-app learning, Resources, and Docs remain available.