All guides

HELP GUIDE · PLATFORM

Connect your AI tools and resolve MCP errors

Check the endpoint, organization, scopes and approval controls behind a failed AI tool request.

Start here

Deepwater’s stateless MCP is a connector to your business tools. The AI client’s identity, organization and allowed scopes determine what it can discover and do. Connecting an AI tool does not grant unrestricted access to all records or modules.

Connect deliberately

  1. Use the MCP endpoint and client configuration supplied for your deployment. Client setup differs between hosted connectors and local tools; a public website URL alone is not the complete configuration.
  2. Connect with the intended human or agent identity and workspace. Review API keys or Agents only if your role permits it.
  3. Begin with a simple read request, such as listing the tools available to that connection. Use the smallest set of scopes needed for the task.

Check and recover

For an authentication error, verify the endpoint and current credential or reconnect the client. Never paste the credential into the question you ask an AI model.

For a permission error, check the selected organization, active membership and scopes. A read-only connection cannot perform writes, and record visibility still applies to linked information.

For missing Signal or Funnel results, confirm that the module is accessible and its organization-bound source connection is configured. An empty result, an unavailable source and an access denial mean different things.

For an action awaiting approval, use the approval workflow. Rephrasing or repeatedly retrying the request does not replace approval, and agents cannot perform every human-admin action.

When to contact the team

Include the AI client, endpoint path, tool name, intended action and sanitized error. Exclude tokens, authorization headers and private records.

Related: Team roles and access.

CHECK THE OUTCOME

Did this get you unstuck?