Create & connect
Use Neurothread from AI assistants
Connect an MCP-compatible assistant to your Neurothread account, choose its access, and manage or revoke that access.
Before you start
Section titled “Before you start”Neurothread can act as an MCP server for an external assistant. This is the opposite direction from Connected apps, where models inside Neurothread call tools on another service. Find the account-level control under Settings → From assistants. The official Neurothread MCP guide is the source for client-specific setup and supported behavior; use it instead of copying a setup from an older article.
An MCP client may connect using OAuth or a personal token, depending on what the client supports. The permissions you grant are account access. Only connect an assistant and device you trust, and treat any token as a password.
When to use this
Section titled “When to use this”Use this integration when you want an external assistant to work with your Neurothread workspaces. The assistant can have read-only access, or read-and-write access if you allow it. Write access lets it create workspaces and ask models on your behalf; answers created this way are saved in your Neurothread workspaces and use your credits. It does not configure external apps for models inside Neurothread.
- Sign in to Neurothread and open the avatar menu → Settings → From assistants.
- If your assistant supports OAuth, follow its connection prompt and review the Neurothread consent screen before allowing access. Choose whether to grant read-only or read-and-write permissions when the client offers that choice.
- For a client without OAuth, create a personal token in the From assistants tab. Give it a recognizable name and leave Allow writing off if the assistant only needs to read. If it needs to create workspaces or ask models, enable writing only after considering the credit and data implications.
- Copy the token when it is shown; it will not be shown again. Store it in the client’s secure configuration using the official MCP setup guide. Do not paste it into a conversation, screenshot, shared document, or support request.
- Use the client’s own connection instructions to finish setup and sign in. Ask the assistant to access only the workspace or information needed for the task, then check the resulting work in Neurothread.
- To remove access, return to Settings → From assistants and select Revoke for the relevant token or authorization. Revoke access you no longer recognize or use.
Worked example
Section titled “Worked example”You want an assistant to find a previous answer but not create anything. Connect it with OAuth and grant read-only access, or create a personal token with Allow writing unchecked. If later you want it to ask a model, revisit permissions deliberately: write access can spend credits, and generated answers are saved in your Neurothread workspace.
What happens next
Section titled “What happens next”The assistant works with the Neurothread account permissions you granted. Access to read your workspaces is distinct from permission to create workspaces or ask models. You can review connected OAuth clients and personal tokens in From assistants, including the access level and recent-use information shown by the interface, and revoke them there. For exact client setup, sign-in steps, and capabilities, follow neurothread.ai/mcp.
Cost and limitations
Section titled “Cost and limitations”Read-only access does not authorize the assistant to create or ask. Write-enabled requests that ask models use your Neurothread credits; the provider and model selected by the request determine the applicable route. A personal token is sensitive: anyone who obtains it may act within its granted scope until you revoke it. OAuth clients may request access through a browser consent flow. This guide intentionally does not reproduce client configuration, endpoint details, or the MCP tool list; those can change and are maintained in the official MCP guide.
Troubleshooting
Section titled “Troubleshooting”- The client asks to sign in: complete the OAuth prompt and review the requested permissions. If OAuth is unavailable, see whether the client supports a personal token.
- The assistant can read but cannot ask: its grant may be read-only. Review the permission in Neurothread before granting write access.
- The assistant cannot connect: reopen the official guide for your exact client and current setup. Avoid reusing old copied configuration.
- A token is lost or exposed: revoke it in Settings → From assistants, then create a replacement if needed. Do not try to recover or share the old secret.
- You meant to let Neurothread models use another service: manage that connection under Settings → Connected apps instead; it is a separate integration.