Connecting your AI agent
Choose your AI client, add brainattic, then try a read-only request. Use the web interface for the setup below; menu labels are shown in English.
One endpoint for every client: remote MCP over Streamable HTTP, with OAuth discovery and automatic client registration. No API key or customer-maintained client secret is needed.
Claude
Personal Pro / Max accounts can add a custom connector. Claude also supports one custom connector on Free. For Team, an Owner or Primary Owner adds it; Enterprise also permits a custom role with Manage access to Libraries.
- For a personal account, open Customize → Connectors → + Add → Add custom connector.
- For Team / Enterprise, the owner opens Organization settings → Connectors → Add → Custom → Web.
- Name it brainattic, paste the server URL above, then Continue through the detected authentication settings. Choose Sign in now and, under OAuth client, Register automatically. Leave manual credentials and Request headers empty, then Add.
- Organization members then open Customize → Connectors → brainattic → Connect and complete their own sign-in.
- In a chat, open + → Connectors and turn brainattic on for that conversation.
Your client opens brainattic’s sign-in and authorization page. Use your own brainattic account, review the requested access, and return to your client after approval. Each person signs in separately; the connection cannot exceed that account’s workspace permissions.
Try the connection
Use Brainattic to list one document in this workspace. Do not change anything.
ChatGPT
Use ChatGPT on the web. Personal and managed workspaces have separate plugin access. In Business / Enterprise / Edu, an administrator controls availability through Workspace apps and Permissions & roles; creation, sharing and publishing permissions can differ. If the add option is missing, check your plan and workspace access with your administrator.
- Open Plugins → + → Add custom MCP server.
- Name it brainattic. Under Connection, enter the server URL above. Select OAuth authentication; use automatic registration (DCR) if asked, without a manual Client ID or Client Secret.
- Review the warning, select I understand and want to continue, then Create as a plugin and complete brainattic authorization when prompted.
- Find the new plugin in your personal plugins or the workspace where you created it. Open it and use + to install it. Sharing a workspace plugin does not install it for every member.
- Start a new Work chat. Type @ in the prompt box and select brainattic, then send the test below. Each member connects their own brainattic account.
Your client opens brainattic’s sign-in and authorization page. Use your own brainattic account, review the requested access, and return to your client after approval. Each person signs in separately; the connection cannot exceed that account’s workspace permissions.
Try the connection
Use Brainattic to list one document in this workspace. Do not change anything.
Mistral
Use Mistral Vibe’s Work mode (formerly Le Chat). A workspace administrator adds custom MCP connectors; on Free / Pro / Student, the account owner is the administrator by default. Vibe Code CLI does not support this OAuth connection.
- In Vibe Work, open Connectors in the sidebar → + Add Connector → Custom MCP Connector.
- Set Connector name to brainattic, paste the full Server URL above, then Connect. Let Mistral discover OAuth and complete the sign-in it opens.
- In your conversation, choose + (or type /) → Tools → brainattic to enable the connector. Review any tool approval before continuing.
Your client opens brainattic’s sign-in and authorization page. Use your own brainattic account, review the requested access, and return to your client after approval. Each person signs in separately; the connection cannot exceed that account’s workspace permissions.
Try the connection
Use Brainattic to list one document in this workspace. Do not change anything.
Other MCP clients
Choose a client that supports remote Streamable HTTP MCP and OAuth discovery with dynamic client registration. Local-only or API-key-only clients cannot use this connection path. Organization policy may require an administrator to add the server.
- Open your client’s MCP server settings and add a remote HTTP server named brainattic, using the server URL above. Menu names depend on the client.
- Select Streamable HTTP and OAuth discovery / automatic registration. Connect and complete brainattic’s authorization; do not substitute No authentication or paste a fixed secret.
- Enable the server’s tools in the conversation or agent if your client has a separate tool switch, then send the test below.
Your client opens brainattic’s sign-in and authorization page. Use your own brainattic account, review the requested access, and return to your client after approval. Each person signs in separately; the connection cannot exceed that account’s workspace permissions.
Try the connection
Use Brainattic to list one document in this workspace. Do not change anything.
How access works
What the connector does
brainattic exposes your workspace to AI agents through the Model Context Protocol (MCP). The kb_* tools let a connected agent list, search, read, create, and edit the documents, projects, groups, attachments, and reminders of the workspaces it is entitled to — the same content your team sees in the web app, under the same permission rules.
Authorization
A connector authorizes with OAuth against our identity provider and receives the brainattic:access scope bound to your account. Fine-grained rights (who may read, write, or delete what) are decided inside brainattic by your workspace's role settings — the connector can never exceed what your account is allowed to do.
Per-tenant isolation
Every token names the workspaces your account belongs to; every tool call is confined to them. Content of other customers is structurally unreachable — isolation is enforced on every query, not by convention.
Read, write, destructive
Read tools (list, get, search) only retrieve. Write tools (create, update, edit sessions) modify content and record every change with authorship and revision history. Destructive tools (delete, archive) are separated and permission-gated, and everything an agent does lands in the workspace audit log — reviewable by your admins at any time.