After deployment, add the Feishu MCP connection in Amazon Quick (Quick Desktop) with the steps below. (For Kiro / Claude Code / Codex, which self-register without a Client Secret, see connect-mcp-clients_en.md instead.)
Before creating the connector, ensure the Redirect URL from deploy output (<OAuth endpoint>/callback) is registered in your app's developer console:
- Feishu (China): https://open.feishu.cn → app Security Settings → Redirect URLs
- Lark (International): https://open.larksuite.com → app Security Settings → Redirect URLs
Use the console that matches your LARKSUITE_CLI_BRAND setting. If this is missing, the authorization step will fail with error code 20029 ("Redirect URL invalid").
In Quick Desktop click Settings → Capabilities → Browse connections (opens browser), choose Create for your team → Model Context Protocol:
If prompted about an existing MCP connector, click No, create new:
Fill in Name, Description, MCP server endpoint (from deploy output), Connection type Public network, click Next:
Fill in Client ID, Client Secret, Token URL, Authorization URL (all from deploy output), click Create and continue:
| Field | Value |
|---|---|
| Client ID | lark-mcp-on-agentcore |
| Client Secret | From deploy output (saved in .local/deploy-output.md), or retrieve with:aws ssm get-parameter --name /lark-mcp-on-agentcore/oauth-client-secret --with-decryption --query 'Parameter.Value' --output text --region <your-region> |
| Token URL | <OAuth endpoint>/token |
| Authorization URL | <OAuth endpoint>/authorize |
A Feishu authorization page automatically opens in the browser, click Authorize:
After authorization, automatically returns to Quick:
Choose visibility (default: only you; or "Everyone in your organization"), click Publish:
After publishing, the Connector detail page shows all available tools:
Back in Quick Desktop, Settings → Capabilities → Connections, search "feishu", click Sign in:
Once connected, Feishu tools are available in conversations.
Once connected, ask Quick Desktop's built-in AI to operate Feishu — it calls this service to reach the Feishu APIs (the natural-language understanding and orchestration are handled by Quick Desktop's own model):
> Check my Feishu calendar for today
> Send a message to the product dev group: sync requirements tomorrow at 3pm
> Summarize last week's meeting notes into a doc
> Add a bug record to the Bitable
Every action runs under the user's own Feishu identity — data is isolated per user.