Tools

Tools Overview

Individual callable functions within toolkits that AI agents can discover and invoke.

What are Tools?

Tools are individual callable functions that live within toolkits. Each tool has a name, description, input schema, and is backed by a serverless function or HTTP endpoint. When an AI agent connects to an MCP server, it discovers the available tools and can call them to perform actions in the real world.

For example, a "Pushover Notifications" toolkit might contain a send-notification tool and a send-priority-alert tool, each with their own input schema and backing function.

Tool Properties

Every tool has the following properties:

PropertyDescription
idUnique identifier (UUID), auto-generated
toolkit_idID of the parent toolkit this tool belongs to
nameUnique string identifier for the tool (e.g., "send-notification")
display_nameHuman-readable name shown in the UI
descriptionA brief description of what the tool does, sent to the AI agent for context
function_nameThe serverless function or HTTP endpoint that backs this tool
input_schemaJSON Schema defining the tool's expected input parameters
requires_credentialsWhether the tool needs credentials to execute (boolean)
credential_provider_nameName of the credential provider to use when resolving credentials
allowed_rolesArray of roles permitted to call this tool (for access control)
dangerousWhether the tool performs destructive or irreversible actions (boolean)
timeout_secondsMaximum execution time before the tool call is aborted
is_activeWhether the tool is currently enabled and available for use
usage_countTotal number of times the tool has been invoked

How Tools Work

When an AI agent calls a tool via the MCP protocol (tools/call), Agent Dojo handles the full execution lifecycle:

1

Tool Resolution

Agent Dojo resolves the tool by name from the server's attached toolkits and validates the input against the tool's schema.

2

Credential Injection

If the tool requires credentials, Agent Dojo resolves them from the linked credential provider and injects them into the function call automatically.

3

Function Invocation

The backing serverless function is invoked with the validated arguments and resolved credentials.

4

MCP Response

The result is wrapped in the MCP response format and returned to the AI agent as a content array.

API Endpoints

Manage and invoke tools using the following API endpoints:

POST/mcp/tools

Register a new tool within a toolkit

POST/mcp/tools/list

List all available tools (MCP-compatible)

POST/mcp/tools/call

Call a tool by name with arguments (MCP-compatible)

Next Steps