# Troubleshooting (/docs/mcp/troubleshooting)



The assistant cannot connect [#the-assistant-cannot-connect]

| What you see                                        | What to do                                                                                                                      |
| --------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------- |
| The connector shows an error right after you add it | Check the URL: `https://feedhorizon.com/mcp`, with `https` and without a trailing path                                          |
| The sign-in page does not open                      | Allow pop-ups for the assistant's site, or start the sign-in again from its connector settings                                  |
| ChatGPT does not offer custom connectors            | Turn on **Developer mode** under **Settings → Apps & Connectors → Advanced settings**                                           |
| A client configured with a JSON file answers `401`  | The `Authorization` header must be `Bearer fh_…` with a key that exists and has not expired. Check the key under **API & Apps** |

The assistant sees the wrong data, or none [#the-assistant-sees-the-wrong-data-or-none]

* **Wrong organization.** An OAuth connection works in the organization you picked on the consent page. Disconnect the app under **API & Apps → Connected apps**, connect again and pick the other organization.
* **Missing profiles.** The assistant sees what you see. If your role is limited to some profiles, so is the assistant; an API key limited to profiles works the same way.
* **No accounts.** The assistant cannot connect social accounts. Connect them under **Connections** in the dashboard, or ask the assistant for a connect link (`create_connect_link`) and open it.
* **It stopped working after you left a team.** A connection belongs to you and the organization. When you are no longer a member, it stops working.

A tool call fails [#a-tool-call-fails]

Errors come back to the assistant as JSON with `code`, `message` and a `hint` it can follow. Usually the assistant fixes the call itself. The common ones:

| Code                                | Cause                                                       | Fix                                                                                                                  |
| ----------------------------------- | ----------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------- |
| `insufficient_scope`                | The connection may only read                                | With OAuth: disconnect and connect again, and allow write access. With an API key: use a read & write key            |
| `resource_not_found`                | An id is wrong or belongs to a profile you cannot see       | Ask the assistant to call `get_context` first; it lists profiles and accounts                                        |
| `validation_failed`, `post_invalid` | The post breaks a rule, such as text that is too long for X | The `details` say what; the assistant can shorten the text or drop an account                                        |
| `queue_not_configured`              | "Add to the queue" on a profile without queue slots         | Set slots (`set_queue_slots`) or give a fixed time                                                                   |
| `rate_limited`                      | Too many calls in one minute                                | Wait the seconds in `retryAfter`. Tool calls count toward your [rate limit](/docs/api/rate-limits) like API requests |
| `internal_error`                    | Something failed on our side                                | Try once more. If it fails again, contact support with the time of the call                                          |

Every code is listed in [Errors](/docs/api/errors).

The post went out at the wrong time [#the-post-went-out-at-the-wrong-time]

Times are in the profile's time zone. Check the profile's time zone in the dashboard (or ask the assistant to call `get_context`, which shows it with the profile's local time), and tell the assistant times as "Thursday 10:00" rather than a UTC time.

The post did not go out [#the-post-did-not-go-out]

Ask the assistant to call `get_post` for it. Each account shows its status and, if it failed, an error with what to do. The most common cause is an account that has to be reconnected (`account_needs_reauth`). After you reconnect it, ask the assistant to retry the post (`retry_post`). [Publishing errors](/docs/api/errors#publishing-errors) lists every cause.

Still stuck [#still-stuck]

Contact [support](/support) with the assistant you use, the time of the failed call and the error the assistant showed.