MCP Tools Not Showing in ChatGPT: What to Check

Symptoms

You connected an MCP-enabled app to ChatGPT, but the expected tool is missing or is not called. These are three different symptoms:

1. The app itself is missing: the app cannot be found in the app list or picker.
2. The app exists but an action is missing: tool scanning or permissions may not expose that action.
3. The action exists but AI does not use it, or execution fails: the conversation context, permissions or external service may be responsible.

Do not assume every missing tool is an MCP connection failure. Identify the symptom before troubleshooting.

Reference date: October 10, 2026. UI labels and features can vary by plan, workspace and rollout.

What to check first

If you are an app user rather than its developer, start here:

1. Check the account and interface. Custom MCP app support depends on plan and workspace settings. Official guidance describes the relevant Web experience.
2. Check app availability. Find the app in ChatGPT’s app settings or your workspace’s approved app catalog.
3. Complete the required connection. If each user must authorize the app, verify its connected state.
4. Select the app explicitly in a new conversation. Choose or mention the intended app.
5. Make a concrete request. For example, ask the selected app to search available documents. Asking only “what can you do?” does not verify that a tool actually ran.

If the interface differs, follow the latest official product documentation.

Main causes

Cause 1: the app is not enabled or published

A developer-mode draft differs from an app published to a workspace. In managed environments, administrators may need to approve or publish the app before other users can access it.

Check: confirm draft versus published state and whether your account is permitted to use the app.

Cause 2: scanning has not completed

When a custom MCP app is configured, ChatGPT retrieves information about the tools offered by the connected MCP Server. Official setup instructions include Scan Tools after setting the endpoint and any required authorization.

Check: inspect the scan result for the expected tool, and look for interrupted authorization or communication errors.

Cause 3: the MCP Server endpoint is wrong or unavailable

An incorrect URL or unavailable Server can prevent tool discovery.

Check: verify the configured endpoint, Server availability and network path. Receiving HTTP 200 from a website homepage does not demonstrate that MCP requests work.

Cause 4: OAuth or permission requirements block access

An unauthenticated session, expired authorization, failed consent or insufficient permissions can block operations.

Check: users should verify their connection state; developers should distinguish authentication failures from discovery and execution failures.

Caution: never paste access tokens, Client Secrets or authorization codes into public tickets or screenshots.

Cause 5: new tools were not reflected in ChatGPT

Adding a Tool to an MCP Server does not guarantee that an already-published ChatGPT app automatically exposes it. Depending on the workspace, an administrator may need to update and review the available tool list.

Check: compare the latest tools advertised by the Server with the actions approved in ChatGPT.

Cause 6: the Tool is available but AI did not choose it

Discovery does not guarantee invocation. The user request, tool description and app selection can affect whether the AI calls a tool.

Check: select the app in a new conversation and request a specific operation that matches the Tool’s purpose.

Diagnostic steps

Developers should check endpoint availability, protocol responses, tool discovery, authorization, app publication and actual execution separately.

Resolution

1. Check Server health. Confirm that the intended MCP endpoint is available.
2. Verify protocol responses. HTTP connectivity alone is not sufficient; test an MCP request supported by the Server.
3. Check tool discovery. Verify that the intended Tools are available. The July 2026 MCP revision introduced stateless connection behavior and server/discover; use tools/list according to the supported protocol version.
4. Separate authenticated from unauthenticated cases. Inspect the behavior before authorization, after authorization and with insufficient privileges.
5. Verify ChatGPT publication and tool settings. Tools that exist on the Server might not be enabled or visible in the ChatGPT app.
6. Run an actual operation. A visible list entry is not proof that a Tool can execute successfully.

Symptom-by-symptom fixes

Symptom	Check first	Suggested action
App missing	Plan, workspace permissions, publication	Verify app availability
No tools from scan	Endpoint, Server health, authorization	Test connection and authentication
Only new tools missing	Stored tool definitions	Review the published tool list
Repeated authorization	Consent, expiration, settings	Review connection status and errors
AI never calls Tool	App selection, prompt, Tool description	Explicitly select the app and request an action
Tool invocation fails	Permissions, inputs, target state	Inspect errors and execution conditions

Distinguishing “not visible”, “not called” and “failed during execution” makes troubleshooting far more precise.

OGINAU implementation observations

OGINAU has tested connections between ChatGPT and external environments while developing OGINAU Remote and OGINAU Browser.

A critical distinction is whether ChatGPT can discover a Tool versus whether a permitted operation actually runs. Even if discovery succeeds, authorization and the target environment may prevent execution. Conversely, an execution failure does not necessarily indicate a discovery defect.

Separating connection checks, Tool discovery, permissions and execution results helps identify the failing stage without exposing proprietary authentication or control internals.

Sources and primary information

• OpenAI Help: Developer mode and MCP apps in ChatGPT: app setup, Scan Tools, permissions, publishing and updates.

• OpenAI Developers: Troubleshooting: MCP Server and tool-discovery diagnostics.

• MCP July 28, 2026 specification announcement: connection, capability discovery and authorization updates.

• OGINAU Remote and OGINAU Browser engineering verification records: separate discovery, access and execution checks.

Consult current official documentation for UI and compatibility changes.

Prevention

After changing an MCP Server, review both the tools discovered from it and the actions approved in ChatGPT. Test the connection, authorization and execution separately.

When adding a new Tool, do not assume an existing app connection automatically updates. Verify app settings and an actual result.

If MCP tools are missing, check app availability first, then Server connectivity and Tool discovery, followed by authentication, permissions and execution. This sequence usually narrows the cause quickly.