MCP Tools Reference

MCP tools are the approved actions an AI assistant can use in Automation Hub.

You can think of them as buttons the AI is allowed to press. Some tools help it find the right automation, some start work, and others check or control a running job.

Tool Groups

Discover

Find the right automation

Run

Start the selected automation

Monitor

Check progress and results

Control

Pause, resume, or stop when needed

Simple: the AI first finds the right job, then runs it, watches it, and reports the outcome.

Where Tools Come From

An agent gets its callable tools from two places in the AI Hub. Both are referenced by id from the AI Agent node.

Source	Node field	What it provides
MCP server	toolsConfig.mcpServers[]	Tools discovered from a registered Model Context Protocol server
Tool instance	toolsConfig.tools[]	A named, pre-configured catalog operation

Tools Discovered from an MCP Server

When you run discovery on a registered MCP server, the server returns its tool set. Each discovered tool is stored as three fields:

Field	Type	Description
name	string	The tool identifier the agent calls
description	string	What the tool does
inputSchema	object	JSON Schema describing the tool's expected input

The agent reads inputSchema to know what arguments to pass. You do not configure these tools individually — they come from the server as discovered.

AI Tool Instances

A tool instance is a named, pre-configured catalog operation. Instead of exposing a raw operation, it binds an operation to a connection and a set of parameters, so the agent invokes a ready-to-run tool.

Field	Type	Notes
name	string	Unique instance name
toolType	string	Stored lowercased on upsert
entity	string	The catalog operation id this tool runs
description	string	What the tool does
configuration	map	connectionId, resource/operation, parameters, and {{...}} expressions
tags	string[]	Free-form labels

The configuration map holds the connectionId (credentials are decrypted at call time), the resource and operation to run, and parameter values. Parameter values may include {{...}} expressions that resolve from workflow data before the call.

Testing a Tool Instance

POST /ai-hub/tools/{id}/test runs an honest test. It looks the operation up in the agent tool-service catalog (roughly 92 operations) to read its destructive flag, then validates the instance's configuration against the operation's input_schema. Validation is tolerant of camelCase and snake_case key differences.

A destructive operation is never executed by the test — it is only validated. Non-destructive operations may be executed to confirm they work.

The response reports what happened:

Field	Description
toolInstanceId	The tested instance
toolType	The instance's tool type
operation	The catalog operation id
destructive	Whether the operation is destructive
executed	Whether the operation actually ran (always false for destructive)
result	SUCCESS or FAILURE
message	Human-readable detail
testedAt	When the test ran

If the catalog cannot be reached, the test treats the operation as destructive: true and falls back to a presence-only check of the configuration.

---

The rest of this page describes the higher-level automation tools an assistant uses to find and run automations.

Tool Summary

Tool	What it lets the AI do
listAutomations	See available automations
searchAutomations	Find automations by name, purpose, tag, or category
getAutomationDetails	Understand what one automation does and what information it needs
executeAutomation	Start an automation
getExecutionStatus	Check whether a run is still running, completed, failed, paused, or stopped
getExecutionResult	Read the final output after a run completes
stopExecution	Stop a running automation
pauseExecution	Pause a running automation
resumeExecution	Continue a paused automation

Discover Tools

listAutomations

Use this when the AI needs to see what automations are available.

Example user request:

"What automations can I run for finance?"

What the AI does:

• Looks through available automations
• May narrow by workspace or collection
• Shows a short list of useful matches

searchAutomations

Use this when the user describes what they want, but may not know the exact automation name.

Example user request:

"Do we have anything for monthly sales reporting?"

What the AI does:

• Searches names, descriptions, tags, and categories
• Finds likely matches
• Asks the user to confirm if more than one option looks relevant

getAutomationDetails

Use this when the AI needs to understand one automation before running it.

Example user request:

"Run the monthly sales report."

What the AI checks:

• What the automation does
• What inputs are required
• What steps it includes
• Whether it has recent successful runs

Run Tool

executeAutomation

Use this when the AI has selected the right automation and has the required information.

Confirm Job

Make sure the selected automation is the right one

Fill Inputs

Use details like month, region, customer, or recipient

Start Run

Automation Hub begins execution and creates a run ID

Example user request:

"Generate the March report and email it to finance."

The AI should confirm missing or risky details before starting the automation.

Monitor Tools

getExecutionStatus

Use this while an automation is running.

The AI can answer:

• Is it still running?
• Which step is active?
• Did it fail?
• Is it paused?
• How far has it progressed?

getExecutionResult

Use this after the automation finishes.

The AI can answer:

• Did it complete successfully?
• What output was created?
• Who received the email or notification?
• What record, file, or report was produced?
• What should the user do next?

Control Tools

stopExecution

Use this when continuing the run would be wrong or risky.

Good reasons to stop:

• Wrong input was used
• User started the wrong automation
• The job is changing data that should not be changed
• The run is taking too long and needs review

pauseExecution

Use this when the job should wait before continuing.

Good reasons to pause:

• Someone needs to review data
• Approval is needed
• A downstream system is temporarily unavailable
• The user wants to verify the next step

resumeExecution

Use this when a paused run is ready to continue.

Good reasons to resume:

• Review is complete
• Approval is given
• The external system is available again
• The user confirms it is safe to continue

A Typical AI Tool Flow

Search

Find the right automation

Inspect

Check required inputs and purpose

Run

Start the automation

Monitor

Watch until it completes or needs attention

Report

Summarize the result for the user

When the AI Should Ask Before Acting

Situation	Recommended behavior
More than one automation matches	Ask the user which one to run
Required input is missing	Ask for the missing value
The action changes important data	Confirm before running
The user asks to stop a job	Confirm which run should be stopped
The automation failed	Explain the failed step and ask before retrying

Common Statuses

Status	What it means
RUNNING	The automation is still working
COMPLETED	The automation finished successfully
FAILED	Something went wrong
PAUSED	The run is waiting before continuing
STOPPED	The run was stopped manually

Good Tool Use Practices

Search before running

• If the user did not name an exact automation, the AI should search first

Check details before execution

• The AI should know what inputs are required before starting a run

Monitor long-running work

• The AI should not say a job is complete until it actually finishes

Explain results simply

• The user usually needs the outcome, not raw execution data

Be careful with control actions

• Stop, pause, and resume should be used deliberately

Related Pages

• AI Integration Overview - the big picture of AI and Automation Hub
• AI Hub Configuration - defining tool instances and MCP servers
• MCP Server - registered servers and tool discovery
• Building AI Agents - guardrail enforcement and human approval