> For the complete documentation index, see [llms.txt](https://mcp-docs.nichefire.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://mcp-docs.nichefire.com/help-and-support/troubleshooting.md). # Troubleshooting This guide covers the most common issues in the order they typically occur: connecting to the server, authenticating, and using the available tools. If your issue is not covered below, see [Contact Us](https://chatgpt.com/c/contact-us). ### Connection issues #### The connector will not save, or the client cannot reach the server Check the following: * **The URL must include `/mcp`.** Use the complete endpoint exactly as provided, not only the hostname. See [Environments & Access](https://chatgpt.com/c/environments-and-access). * **Your client must support remote MCP servers.** Nichefire MCP is hosted remotely. Clients that only launch MCP servers locally cannot connect to it. See [Prerequisites](https://chatgpt.com/c/prerequisites). * **Outbound HTTPS may be blocked.** This generally affects desktop applications and code editors on managed networks. It does not normally affect cloud clients such as Claude.ai or ChatGPT, which connect from their own infrastructure. If you are using Claude Desktop, Claude Code, Cursor, or VS Code on a managed network, ask your IT team to allowlist the Nichefire host. #### The connector saves, but no tools appear This usually means the connection and authentication succeeded, but your organization has not been granted access to any capability family. Contact your Nichefire representative to confirm your organization’s access. #### Some tools appear, but others do not This is expected. Clients display only the tools included in your organization’s entitlements. For example, if you expected Trend Insights but can only see research tools—or the reverse—this is an access question rather than a connector misconfiguration. See [Capabilities](https://chatgpt.com/c/capabilities). #### The client asked me to sign in, but sign-in failed Possible causes include: * **Your account is not associated with an organization that has MCP access.** Signing in verifies your identity, while MCP access is granted at the organization level. Contact your Nichefire representative. * **You signed in with the wrong account.** If you use multiple accounts, make sure you select the one associated with your Nichefire organization. ### Authentication issues #### Missing `x-api-key` header The client did not include your API key in the request. Check the following: * Confirm that the key is configured in the connector’s custom headers. * Some clients require you to save the connector before headers take effect. Reopen the configuration and confirm that the header is still present. * The header name must be `x-api-key`, in lowercase. * The complete API key must be used as the header value, without a `Bearer` prefix. Claude.ai does not provide a custom headers field and supports OAuth authentication only. See [Claude Connectors](https://chatgpt.com/c/claude-connectors). #### Invalid API key For security reasons, this message does not identify the specific cause. Check the following in order: 1. **Confirm that the key is complete.** Copying and pasting can omit characters or add whitespace. 2. **Confirm that the key matches the environment.** For example, a sandbox key cannot be used against the production endpoint. See [Environments & Access](https://chatgpt.com/c/environments-and-access). 3. **Confirm that it is a Nichefire MCP key.** Credentials issued for other Nichefire products cannot be used with the MCP server. 4. **Confirm your organization’s entitlements.** A key cannot call tools from a capability family that has not been enabled for your organization. #### It worked previously, but now asks me to sign in again Stored OAuth credentials can expire or be revoked. Sign in again to restore the connection. This does not usually indicate a problem with your local configuration. #### I reconnected and now see different tools When you approve an OAuth connection, it uses your organization’s most recently issued credential. If a new credential was issued with different entitlements, reconnecting may change the tools available to you. See [OAuth 2.1](https://chatgpt.com/c/oauth-2.1). ### Tool and usage issues #### `Rate limit exceeded` continues after retrying The same HTTP `429` response can represent two different conditions: * **`reason: rate_limit_exceeded`** — Too many calls were made within the current minute. Wait up to one minute and try again. * **`reason: quota_exceeded`** — Your organization has used its research allowance. Waiting will not restore access. Contact your Nichefire representative to request additional quota. You can ask your assistant to call `get_usage` to identify the cause. This tool does not consume quota and remains available while usage is limited. See [Errors](https://chatgpt.com/c/errors). #### My quota decreased by more than the number of questions I asked Quota is measured by tool calls rather than chat messages. A single question may result in multiple calls. Keep the following in mind: * Each search consumes one unit. * A request to compare several topics or perspectives may trigger multiple searches. * Quota is shared across the organization. Calls made by colleagues, scripts, or other connected clients use the same allowance. * `analyze_trend` draws from the same pool and consumes one unit per call. See [Quota](https://chatgpt.com/c/quota). #### A search returned no trends but still consumed quota This is expected. The research was completed, but no distinct trend was identified for the provided prompt. The response includes a reasoning block describing what was searched and why no trend was formed. In many cases, using a narrower and more specific prompt will produce better results. See [`execute_search`](https://chatgpt.com/c/execute-search). #### My in-depth request returned a fast, limited response in ChatGPT Within ChatGPT, searches usually run in responsive instant mode. The extended thinking mode is not used automatically. For longer-form analysis, use a supported client or integration that can explicitly request the deeper research mode. See [Capabilities](https://chatgpt.com/c/capabilities). #### Results differ between ChatGPT and Claude Each client receives the same structured data but may present it differently. For example, one client may display an interactive widget while another provides a written summary. If the underlying content differs, remember that live research uses current web data. Separate runs of the same query may therefore surface different sources or coverage. #### `Prompt must be a valid string` or `Only prompt and researchMode are allowed` The request body is invalid. This is more common in custom integrations than in conversational clients. Make sure that: * `prompt` is a non-empty string. * The request contains no fields other than `prompt` and `researchMode`. See [`execute_search`](https://chatgpt.com/c/execute-search). #### The assistant cannot find a topic or trend that I know exists Trend Insights can only access saved data belonging to your organization. Content stored in another organization’s workspace is not visible to your credential. Trends are also addressed by ID. The assistant may need to list your available topics before it can locate the relevant trend. Ask it to list your trend topics and continue from there. See [`list_trend_topics`](https://chatgpt.com/c/list-trend-topics). #### `Internal server error` or `Failed to parse research output` The request failed on the Nichefire side. Retry the operation; any quota unit consumed by the failed request is refunded automatically. If the problem continues, contact [Nichefire Support](mailto:support@nichefire.com) and include: * The approximate date and time of the request * The client or integration you were using * The prompt or action you submitted * The complete error message * Any other details that may help reproduce the issue