Skip to main content

Tools

Estimated Time: 2 minutes

Tools are capabilities shared by agent skills and MCP servers. They let agents search your knowledge base, call your HTTP endpoints, connect to supported apps, or exchange structured data with the website where the chatbot is embedded.

You add tools from Knowledge and actions in the agent builder or the MCP server's Tools tab. Each tool has a name, a short description that helps the assistant know when to use it, and the configuration needed to run it.

Available Tool Types​

Knowledge Base & RAG​

Use uploaded documents, FAQs, policies, product information, or other files as a searchable knowledge base. Predictable Dialogs processes the files into chunks, stores them in a vector store, and retrieves relevant chunks when the skill needs document-backed answers.

Learn more: How Knowledge Base & RAG works

HTTP Tool​

Call your own API endpoints with GET, POST, PUT, PATCH, or DELETE. Use HTTP tools to look up live data, create records, update systems, or trigger workflows through your backend.

Learn more: HTTP Tool

App Connection Tool​

Connect a supported app account and choose which app actions the skill can use. Google Calendar is currently supported for searching available slots and booking appointments.

Learn more: App Connection Tool

Send Data to UI​

This is a website agent tool.

Send structured JSON from the assistant to your website. Use agentUi.components to show it inside the conversation, or onToolResult to update UI elsewhere on the page after the response finishes.

Learn more: Send Data to UI

Request User Input​

This is a website agent tool.

Ask a visitor for information through custom interactive UI inside the conversation. The assistant provides the UI definition; your registered renderer displays it, and the widget submits or cancels the visitor's response.

Learn more: Request User Input

How Tools Run​

For agents, the assistant decides when to use a tool based on the skill instructions, the tool name, and the tool description. You can also configure advanced rules for when a tool is eligible, whether it should run after another item, or whether it should be skipped for some messages.

For MCP servers, the external client chooses which tools to call. Agent-specific execution rules and website UI tools do not apply.

Inspect inputs and results in Sessions for agents or Activity for MCP servers.

Getting Started​

  1. Open your agent's skill or your MCP server's Tools tab.
  2. Go to Knowledge and actions.
  3. Add a Knowledge item or an Action item.
  4. Name the tool and describe when the assistant should use it.
  5. Configure the tool, test it when testing is available, and review tool usage in Sessions or MCP Activity.

Best Practices​

  • Keep each tool focused on one job.
  • Use clear tool names and usage descriptions.
  • Keep input schemas as small as practical.
  • For web agents, use pass-through auth for user-specific app actions whenever the action should run with the signed-in user's permissions.
  • Review Sessions or MCP Activity after testing so you can see the exact tool inputs and outputs.

Next Steps​