FAQ: Use your own AI agent instead of Quackback AI
Connect your own AI agent through the MCP server, webhooks and REST API, give Quackback AI your own tools as connectors, or run it on your own model when self-hosting.
Yes, you can pause the built-in AI agent and connect your own AI agent to Quackback through the MCP server, the REST API and webhooks. Your agent can read conversations, reply to customers, open tickets and triage feedback with the same permissions as an API key you control. This FAQ shows how the pieces fit together.
Option 1: Your own agent through MCP
Quackback includes an MCP server that any MCP-capable agent or AI tool can use.
- Pause the built-in AI agent in Admin > Settings > Agent (Pause Agent) so the two agents don't both reply.
- Turn on the MCP server in Admin > Settings > Developers, on the MCP tab.
- Create an API key on the Keys tab of the same page, and give it only the access your agent needs (for example Conversations: Read and write).
- Point your agent at your workspace's MCP endpoint,
https://<your workspace domain>/api/mcp, with the key as a Bearer token.
Your agent can then use tools such as list_conversations, get_conversation, reply_to_conversation, create_ticket and search. See MCP Server and the MCP tool reference.
Note: On Quackback Cloud, the MCP server and webhooks are included from the Pro plan.
Option 2: React to events with webhooks and the API
For an agent that should respond as soon as something happens:
- Create a webhook in Admin > Settings > Developers > Webhooks for events such as
conversation.createdandmessage.created. - When your service receives an event, have your agent decide what to do.
- Write back through the REST API or the MCP server (reply, set status, open a ticket).
See Webhooks and the API overview.
Option 3: Keep Quackback AI, add your own tools
If you like the AI agent's answers but need it to reach your own systems (orders, accounts, internal docs), connect your MCP server as a connector in Admin > Settings > Connectors. Quackback AI can then call your tools, with Allow, Ask or Never per tool. See Give Quackback AI actions.
Use your own model with Quackback AI (self-hosted)
On a self-hosted install, Quackback AI runs on whatever OpenAI-compatible endpoint you configure, including a local model server. Set OPENAI_BASE_URL to your server and choose models with AI_CHAT_MODEL (or AI_ASSISTANT_MODEL for the AI agent and Copilot only). See Configure AI features. On Cloud, Quackback AI runs on a model Quackback provides.
Related articles
Was this helpful?
Your feedback shapes what we write next.
