Written guide
Companion source code
Get the course source code on GitHub. The folder includes the supplied examples and setup instructions. Keep credentials in your own environment; use the maintained API reference when adapting recorded request examples.
Connect a supported AI client to AppEngine, inspect its available tools, and verify the connection with one read-only question. The example uses Codex CLI; another HTTP MCP client needs the same AppEngine URL and headers in its own settings.
Video status: the earlier extended approval course is withdrawn. The concise 2:26 narrated replacement is ready for owner review. Watch the connection course. Producer and independent scoped checks passed; owner acceptance is pending.
1. Find the maintained connection reference
Open AppEngine documentation, then MCP server in its navigation. The MCP reference is the place to check current connection values and tool behavior.
The hosted MCP endpoint is https://appengine.appmint.io/mcp. Choose Streamable HTTP when your client asks for transport. For a self-hosted AppEngine deployment, use its base URL followed by /mcp. The current endpoint accepts protocol messages by HTTP POST; a browser GET returning 405 does not prove the connection is broken, and an SSE-only client is not compatible with this endpoint.
Documentation publication note: the corrected reference used in the recording is a local documentation preview. Public tool discovery was verified on 11 October 2026, but publication of these documentation corrections is still pending. The old hosted page's “not deployed” statement is stale.
Actual generated documentation preview. It is not represented as an already-published correction.
2. Have the right organization and identity
Obtain the intended organization ID and a staff access token before connecting. API Foundations — sign in as the owner shows the staff sign-in and completed session fields; the maintained staff authentication reference supplies current details. Complete any offered password-change or verification challenge before using the returned access token. Start with Appmint Studio if you still need the platform's sign-in entry point.
The AI client's own login and the AppEngine staff token serve different purposes. Your AI-client login runs the assistant; the AppEngine token authorizes its business-record calls. Use a staff identity from the intended organization. Keep the token private, outside prompts, screenshots and source control.
3. Add AppEngine to your AI client
Have Codex installed and signed in before this step; its official MCP setup guide describes current client setup. For Codex, add the following table to your private ~/.codex/config.toml. Replace the organization value and use your deployment's URL if it differs from the hosted endpoint. Preserve other existing configuration.
[mcp_servers.appengine]
url = "https://appengine.appmint.io/mcp"
bearer_token_env_var = "APPENGINE_STAFF_TOKEN"
http_headers = { orgid = "YOUR_ORGANIZATION_ID" }Start Codex with APPENGINE_STAFF_TOKEN available in its environment. The value is the token itself; the client constructs the Bearer authorization header. This example stores the variable's name, not the token, in the configuration. Other HTTP MCP clients may provide a secrets form or custom-header fields instead.
Actual local editor saving a standalone configuration example. The recording deliberately does not overwrite personal settings or expose a token. Use the entry in your own client's configuration.
After saving, run codex mcp get appengine to inspect the configured entry. Then start a fresh codex session from the environment containing the token; an already-running client may not have the newly exported value. A saved configuration is not proof of a successful connection. Inspect the connected MCP server next.
4. Check the tools and ask one useful question
Open the client's MCP view. In the recorded Codex version, /mcp shows AppEngine as connected with three tools. Their names are list_services, describe_service, and call_service; the subsequent tool invocations make those names visible. These let the assistant discover services, inspect the operation it needs, and invoke it. You do not need to hand-write MCP protocol requests.
Actual installed Codex client, connected to the isolated training AppEngine environment.
Ask a small question about one record you are allowed to read. For example:
Read employee YOUR_EMPLOYEE_RECORD_ID and tell me their name and staff code. Do not change anything.
Obtain that ID from an employee record you may inspect: the maintained API reference lists GET /business-made/employees. Select your intended employee and use its returned sk, not the human staff code in data.employeeId. That staff route requires the appropriate BusinessMade area access. Alternatively, ask the connected assistant to find your uniquely identified training employee first, then verify its name and staff code before using the returned ID. Do not reuse the recording's fixture ID in another organization. In the recording, the assistant discovers the employee service, inspects its contract, calls getEmployee, and answers Tom Becker — mcp-usecase-1011-EMP0042. That is a real tool-backed answer from the synthetic training record, not a manually composed response.
The successful authenticated read uses an isolated training backend. The hosted endpoint was separately checked for public tool discovery; this demonstration does not claim the synthetic record exists in the public service.
5. Know the important boundaries
Discovery can work without a staff token. Record calls require organization context, valid staff authentication and content read permission. If tools appear but a record call fails, check those values and whether the token has expired.
The current MCP gateway calls registered services directly; it does not repeat every REST controller's menu or operation guard. Service-level checks still apply. call_service can invoke reads and writes, so inspect what your client allows. A client confirmation prompt, or “do not change anything” in a prompt, is not a server-enforced permission policy.
If no tools appear, first check the endpoint and HTTP transport against the maintained reference. If the client cannot obtain its token, check the environment from which it was started. These are connection problems; repeatedly asking the assistant a different business question will not repair them.
Recording scope
This concise lesson teaches connecting an AI client to AppEngine. It does not require a timesheet-approval exercise or teach HTTP/JSON programming. The prior extended approval tests remain preserved as production evidence, outside this lesson's main path. Configuration formats and reference details may change; return to the maintained documentation and your installed client's help when setting up a later version.