Problem: Currently, MCP access is an "all or nothing" authorization. There is no way within Hospitable to restrict which specific tools or data an AI agent can access via the Model Context Protocol.
Current Workaround: Users must rely on the AI tool's own internal settings or guardrails, if available, to limit the agent's actions.
Impact: This creates a security and operational risk, as users cannot precisely control the scope of an AI agent's capabilities within their account.
Proposed Solution: Implement granular permission toggles within Hospitable for MCP, allowing users to enable or disable specific MCP tools/scopes for connected AI agents.
Unlock: Users will have peace of mind and tighter security, knowing exactly what an AI agent can and cannot do in their account.