Tools let your Agents and Flows take action — whether that means searching the web, calling an API, running custom code, or connecting to external services. Instead of only generating text, your AI can do work.
Runtype supports several tool types, so you can connect your workflows to almost anything they need.
These ready-to-use tools are provided by Runtype. They appear automatically when you add tools to a prompt step or Agent and select a compatible model. For a deeper overview, see Built-in tools.
Firecrawl and Exa API keys are configured under Settings → Integrations. For GPT Image 2 and other provider tools, set up the provider under Settings → Models. Semantic Search works with any knowledge base you have created in the platform, with no extra setup.
Custom tools let you write JavaScript, TypeScript, or Python code for specialized tasks. They run in Runtype’s sandboxed environment and include built-in helper functions for text processing, date handling, validation, and encoding.
Great for: Data transformations, calculations, input validation, formatting, and business logic.
Custom tools include helpers like parseHTML(), extractEmails(), and formatDate(), so you do not need to build everything from scratch.
External tools connect your Agents and Flows to any HTTP API endpoint. They support multiple authentication methods, including Bearer tokens, API keys, and custom headers.
Great for: Third-party APIs, internal microservices, databases, webhooks, and legacy systems.
Flow tools let you turn any Flow into a reusable tool that other Agents and Flows can call. This is a powerful way to compose multi-step workflows and reuse them across your projects.
Great for: Multi-step operations, orchestrated tasks, and reusable workflows that you want to share across multiple Agents.
If you keep rebuilding the same logic in multiple Flows, extract it into its own Flow and use it as a Flow tool instead.
Runtime tools are defined inline with API requests, such as when you call the dispatch API, without saving them to the platform first. They support the same types as saved tools: external HTTP, custom code, Flow, and subagent.
Great for: Dynamic or temporary tool configurations, SDK-driven Flows, testing, or cases where you do not want to store tool definitions in the platform.
You can connect MCP (Model Context Protocol) servers to expose additional tools to your Agents and Flows. MCP gives AI models a standard way to interact with external services, and Runtype can connect to any MCP-compatible server by URL with automatic tool discovery. To use MCP with a Product, see Setting Up an MCP Surface.
Great for: Extending your Agents with standardized integrations, connecting to internal tools, and using the growing MCP ecosystem.
When you add tools to an Agent, the AI decides when and how to use them.
This happens automatically. Your job is to make the right tools available and write clear descriptions so the Agent knows when to use each one. If you are setting up an Agent from scratch, see Creating and configuring Agents.
Good tool descriptions improve results. Be specific about what the tool does and when the Agent should use it.
You can create and manage saved tools from the Tools section in the sidebar.
Built-in tools do not appear here. They become available when you configure them in Settings and add tools to an Agent or Flow step.