You can do the same from the dashboard with the Agent
Quickstart
Connect your AI assistant to your Attlas workspace in three steps:1
Create an API key
- In the Attlas dashboard, open Workspace settings.
- Go to the API keys (or MCP) tab.
- Click Create API key.
- Enter a name (e.g.
Cursor assistantorClaude Code). - Choose your permissions (Write to create and edit resources, or Delete if you require deletion privileges).
- Click Create key and copy your secret key (
atk_...) immediately.
2
Configure your AI client
The workspace MCP endpoint is hosted at:Select your AI tool below to copy the setup command or configuration file:
3
Verify the connection
Test your setup by asking your assistant:Your assistant will call the
list_chats tool and return the list of chats with their numeric identifiers.Claude web and desktop (custom connector)
To connect from claude.ai or the Claude desktop app, add Attlas as a custom connector. The workspace endpoint authenticates with an API key, not OAuth, so a few fields need specific values.1
Open the custom connector dialog
In Claude, go to Settings → Connectors → Add custom connector (Team and Enterprise: Organization settings → Connectors → Add → Custom).
2
Enter the URL
3
Set Authentication to None
After you continue, Claude probes the URL and may pre-select Always required, marked Detected. This is a false positive: the server returns
401 to the unauthenticated probe, and Claude reads that as OAuth.Select None instead. The description on that option says it explicitly: “or for servers that use an API key rather than OAuth.” The OAuth client section then disappears.4
Add the key as a request header
Open Request headers and add one header:
5
Add the connector
Click Add, then enable Attlas from the + menu in a conversation.
Request headers is a beta feature available to a limited set of organizations. If you don’t see that section, connect with Claude Code or
mcp-remote instead (see Quickstart above).Troubleshooting
Permission scopes
When creating an API key, select the permissions that match your use case:Plan requirements
The workspace MCP server itself works on every plan. A few actions follow the same plan limits as the dashboard:Available tools
Attlas provides 67 tools in total across read, write, and delete scopes. Each tool needs the scope shown next to it, and a key only sees the tools its scope allows.Chats and setup
Knowledge
Adding or editing knowledge takes a moment. These tools answer right away with a
processing status, and your assistant checks list_documents until the document is ready.Triggers
Forms
Campaigns
Contacts and appointments
Insights
Plan
Delete
What stays in the dashboard
Some settings need you at attlas.so. Your assistant tells you when a request falls here:- Changing a chat’s URL.
- Connecting an integration: Google Calendar, Notion, a SQL database, Bandsintown, the Meta Conversions API, a custom domain, or a webhook. Once connected, your assistant can configure and use it.
- Access mode, password, and public or private visibility.
- Plan, billing, and add-ons.
- Workspace members and invitations.
Monitoring activity and audit logs
The Recent activity table at the bottom of the API keys tab logs every MCP call made to your workspace:- When: Relative timestamp of the request.
- Key: Prefix of the key used (
atk_...). - Tool: The MCP tool called (e.g.
list_chats,create_document,delete_trigger). - Target: The resource touched by the operation.
- Result: Success (
ok) or error badge.

