The Tool Studio Assistant helps turn a need into a tool that is ready for your review. It is available while creating tools and inside each HTTP group's management page.
For example, you can ask:
- “I need to look up customers in my CRM.”
- “Find an MCP option for Slack.”
- “Explain how to import this OpenAPI specification.”
- “Check why this tool returns 401.”
- Or paste a
curlcommand directly.
Create from natural language or cURL
Describe the outcome you want and the assistant will recommend MCP, OpenAPI, or manual HTTP creation as appropriate.
When you paste a curl, it can detect and preload:
- the method and URL;
- path, query, and body parameters;
- public and secret headers;
- JSON or
form-urlencodedbodies; - basic authentication or an API key when the structure is recognizable.
The configuration first appears as a proposal. Nothing is saved or changed automatically.
Review and approve changes
Each proposal shows the previous state and the complete change. You can:
- Approve and apply: place every proposed change in the form.
- Discard: keep the previous state.
Approval is atomic: individual fields from the same proposal are never applied separately. You can then review the normal form and choose Create tool, Test tool, or ask for more help.
The same mechanism applies when editing or repairing an existing tool.
Credentials and sensitive data
Credentials entered in chat are detected and replaced before the message is sent to the model. Their values stay temporarily in your browser's memory so the assistant can complete sensitive form fields.
- The model never receives the credential value.
- History stores only a hidden reference, never the secret.
- After a page reload, the chat remains but pending credentials must be entered again.
- A secret is stored only when you confirm the corresponding form, using Brain's encrypted storage.
Find MCP servers online
You can ask the assistant to search for MCP alternatives for a service. Suggestions include links and a provenance label:
| Label | Meaning |
|---|---|
| Public documentation | Information published by Brain. |
| Official | A provider resource or a verified catalog option. |
| Community | A community project on a known public source. |
| Unverified | A public result that you should evaluate before use. |
You can use this information immediately; it does not require manual review by the Brain team. Before connecting an external server, check who maintains it, which permissions it requests, how it stores data, and whether its URL is actually an MCP endpoint.
A catalog suggestion can open its connection flow directly after you authorize it. For any other server, the assistant guides you to Tool Studio → MCP Server, where you confirm the URL and credentials.
Chats and history
You can keep several chats at once, such as one for Outlook and another for Slack. Each chat keeps its messages, sources, and proposals across page reloads. You can archive it when it is no longer needed.
Chats are independent, but always use the context of the current screen. To edit or repair a specific tool, first open the HTTP group that contains it.
Test and assign the tool
After you apply a proposal, the assistant can offer actions such as:
- create or save the tool;
- open its test panel;
- connect an MCP integration;
- open the digital-human selector;
- continue with guided help.
Write tests make a real request. Brain asks for confirmation and does not change the arguments or automatically resend a failed write. If credential renewal is configured and the API returns 401/403, Brain may obtain a fresh credential and repeat the same request once. Read tests may correct arguments and retry within a fixed limit.
Global permissions
A tool's policy is global for your account. If you ask the assistant to change it to Always allow, Require approval, or Block, the proposal affects every digital human that uses the tool, including future assignments.
Assignment remains individual: you choose which digital humans can access the connection. See Permissions and access for the distinction.