Tools covers everything you build in Tool Studio to connect the Moveworks AI Assistant to your business systems and content: the connectors that hold a system’s credentials, the plugins and MCP servers the assistant calls, the actions and system triggers that automate work in the background, and the logs you read once a tool is live. Moveworks offers four tool frameworks, each optimized for a different kind of connection: Search to find content, Plugins for reliable write actions that respect business rules, MCP for live reads and reversible actions in productivity apps, Agent-to-Agent (A2A) to delegate work to other agents.

Build your first plugin in minutes against the Purple Suite mock APIs.

Search, Plugins, MCP, and A2A, and when to use each.

Cookbooks

Collection of best practices and recommendations for common types of use cases.

New to Tool Studio? Start Here

Set up the Purple Suite

Every quickstart uses the Purple Suite mock APIs. Complete Purple Suite Setup once to create your instance and connector.

Build your first plugin

Follow Quickstart #1: Build Your First Plugin to build a working plugin in minutes.

Understand the architecture

Read Core Concepts, starting with Assistants, Agents, & Plugins, to understand how the pieces fit together, then Choosing the Right Tool Framework before you pick a framework for a real use case.

Learn the Golden Rule

Read The Golden Rule before building anything complex — it’s the most important design pattern.

Build with the reference below

Use the By Job and Quick Reference sections on this page to find specific docs as you build.

Test before you ship

Review Testing & Error Handling and the Plugin Go-Live Checklist before publishing your plugin.

Pick a Framework

Match the framework to the kind of work the assistant needs to do. An assistant can use several frameworks at once: when users chat, the assistant decides which tool to use to search for information, analyze data, or take action.

Find content and files across your systems, such as knowledge base articles and attachments.

Plugins

Reliable, agent-safe write actions against enterprise APIs, with step order and business rules enforced.

MCP Workspace

Live reads and reversible operations on productivity apps through a connected MCP server.

Agent-to-Agent (A2A)

Delegate open-ended tasks to other agents and existing AI investments.

When no built-in search connector covers a source, build one with Content Gateway, starting from its Starter Code. Plugins that users activate by chatting are conversational plugins; plugins activated by system events are ambient agents. For the side-by-side comparison, read Choosing the Right Tool Framework. Productivity Assistants on the Agentic Orchestrator use the same plugins, MCP servers, and A2A agents; Tool Orchestration explains how the Orchestrator exposes them.

By Job

Build your first tool

Quickstart #1 walks through a working plugin end to end.

Connect a system

Create a connector that holds a system’s base URL and authentication; every HTTP action inherits it.

Automate on events or schedules

Pair actions with webhook or scheduled system triggers to build ambient agents that run without a user prompt.

Let ToolSmith build it

Describe the workflow, review the proposed plan, and ToolSmith builds or edits the plugin for you.

Test and debug

Launch to yourself, read the logs, and work through common errors before publishing.

Watch it run

Execution logs, MCP tool call logs, assistant analytics, and the Data API once a tool is live.

Event-driven and scheduled plugins follow the ambient agent pattern; Ambient Agents explains the Listen → Understand → Act cycle. Systems behind your firewall connect through the Moveworks On-Prem Agent. If another system needs to call Moveworks APIs, create its credentials under API Credentials.

Quick Reference

Starting from an outcome and not sure which primitive to use? Start here.

I want to…	Use
Call an API	HTTP Action
Run custom code (APIthon)	Script Action
Orchestrate multiple actions	Compound Action
Call an LLM inline	generate_text_action / generate_structured_value_action
Send a message to a user	`notify` expression
Branch on a condition	`switch` expression
Handle an API failure	`try_catch` / `raise`
Loop over a list	`for` expression
Run actions concurrently	`parallel` expression
End a compound action and return data	`return` expression
Look up a user by email	`get_user_by_email`
Look up many users at once	`batch_get_users_by_email`
Request human approval	`create_generic_approval_request`
Send batch chat notifications	`batch_send_plaintext_chat_notification`
Collect user input	Slots
Validate a slot value	Validation policy with DSL
Write a condition or validation rule	DSL Reference
Reshape an API response	Data Mappers: MAP(), MERGE(), FILTER(), RENDER(), LOOKUP()
Branch a conversation on an action result	Decision Policy
Display a knowledge article or form	Content Activity
Handle large API responses (>7K tokens)	Response Sizes / SDA

Actions

HTTP, Script, Compound, LLM, and Built-in actions

Conversation Process

Slots, resolver strategies, activities, and control flow

Natural Language Triggers

Conversational plugin activation

System Triggers

Webhooks and scheduled triggers

Data Bank

Reference patterns: data., meta_info., response, value

Data Types

Built-in and custom data types

Guides & Best Practices

The Golden Rule

The single most important pattern for building reliable plugins

Decision Frameworks

When to use compound actions vs conversation processes, slot types, and more

Plugin Building Masterclass

Build a plugin end to end, from the downstream API through launch, testing, and operation

Slot Best Practices

Anti-patterns and best practices for slot design

Conversation Process Best Practices

Anti-patterns and best practices for conversation processes

LLM Action Best Practices

When NOT to use LLM actions and the decision matrix

The Guides tab also covers Testing & Error Handling, Pass Data to Dynamic Resolvers with Strategy Mapping, Access Control Best Practices, Building a Conversational Plugin, Building an Ambient Agent, Choose How to Connect External Systems to Moveworks, and LLM Orchestration Patterns.

Samples

You can view over 100 pre-built AI agents on our AI Agent Marketplace and import example templates to better understand how they are built.

How Tools Fits With the Rest of the Platform

Tools describes what the assistant can call. The Harness explains how the Agentic Reasoning Engine decides which of your tools to use in a conversation, and Why Plugins explains how plugins follow your business rules every time once selected. Who can launch and use each plugin or MCP server is set through launch configuration in Platform Management.