Table of Contents

Tools

Learn how to manage the tool library, tool families, and tool sets, and how to attach tools to an agent.

Overview

The Tools page is the catalog of tools that agents can use. It holds two kinds of tools:

  • Built-in tools ship with the platform (for example Code, Knowledge, Perplexity Search, Gemini Image Generation, and Claude Agent).
  • Organization tools are tools your team creates, such as an MCP tool that calls your own MCP server.

Tools are grouped into tool families (for example Perplexity, Google, or Anthropic) and can be bundled into tool sets so a group of tools can be added to an agent in one step.

When you add a tool to an agent, the portal copies the library tool into the agent. Later edits to the library tool change the defaults for new assignments only; existing agents keep their copy.

To open the page, in the sidebar select Agents > Tools. The page has two tabs:

Tab Purpose
Tool Library Browse, search, create, edit, and delete tools and tool families
Tool Sets Create and maintain bundles of tools

You can link directly to a tab with /tools?tab=library or /tools?tab=sets.

Tool Library

The Tool Library tab shows one collapsible folder per tool family. Folders are collapsed by default. Each folder header shows the family icon, the family display name, and the number of tools in it.

Browsing and searching

  1. Select a folder header to expand or collapse it. Use Expand All or Collapse All to change every folder at once.
  2. Type in Search tools and sets... to filter. The search matches the tool display name, description, resource name, and family name. Folders that contain matches expand automatically. Select the × icon in the box to clear the search.
  3. Use the origin filter to show All origins, Built-in, or Organization tools.

Each tool card shows the display name, its origin (Built-in or Organization), and its description. Cards have two actions:

Icon Action
Pencil Opens the tool editor
Trash Deletes the tool after confirmation

Tools that are not assigned to any family appear in an Other folder.

Note: Tools, tool families, and tool sets created or updated by an artifact package cannot be deleted from the Management Portal. For those resources the delete icon is replaced by an information icon that explains this.

Tool families

A tool family groups related tools and supplies the vendor icon shown on its tools. Families do not hold endpoint or model settings.

Create a family

  1. On the Tool Library tab, select Create Tool Family.
  2. In the Create Tool Family dialog, enter a Display name.
  3. Review the Name. The portal derives it from the display name and checks that it is available (✔️ available, ❌ taken or invalid).
  4. Optionally enter a Description and choose a Vendor icon.
  5. Select Create family.

Edit a family

  1. Select the pencil icon on the family folder header.
  2. In the Edit Tool Family dialog, change the Display name, Description, or Vendor icon. The resource name cannot be changed.
  3. Under Tools in family, select Add Tool to move a library tool into this family, or select the trash icon on a row to remove it.
  4. Select Save family.

Delete a family

A family can only be deleted after all of its tools have been removed from it. When the family is empty, select the trash icon on the folder header and confirm.

The built-in families are:

Family Resource name Tools
FoundationaLLM Platform-ToolFamily-Basic Code, Knowledge, MCP Client, Programmable Tool Calling
Perplexity Platform-ToolFamily-Perplexity Perplexity Search, Ask, Reason, Research
Google Platform-ToolFamily-Google-Gemini Gemini Image Generation, Gemini Web Grounding
OpenAI Platform-ToolFamily-OpenAI-Image GPT Image Generation
Anthropic Platform-ToolFamily-Anthropic Claude Agent
Cursor Platform-ToolFamily-Cursor Cursor Cloud Agents MCP tools
Genesis Platform-ToolFamily-Genesis Agent lifecycle management MCP tools
Data Platform-ToolFamily-Data PostgreSQL Execute SQL, Search Database Objects

New tools without a family are assigned to FoundationaLLM (Platform-ToolFamily-Basic).

Creating a Tool

  1. On the Tool Library tab, select Create Tool.
  2. In the Create Tool dialog, choose a type and select Next:
Type Use when
MCP Tool The tool calls a tool on an MCP server, directly or through Intelligent Execute
Custom Tool (Advanced) The tool is a class in a FoundationaLLM Python plugin package

MCP tool

The MCP wizard has three steps: Type, Configure, and Review.

  1. On the Configure step, fill in the tool editor (see Tool Editor Reference):
    • Display name, Name, Description, and Tool family.
    • Under Connection, choose Registered endpoint and select an MCP Server Endpoint, or choose Inline URL and enter the MCP Server URL.
    • Under Tool Selection and Configuration, either select List Tools, pick a discovered tool, and select Configure, or enter the MCP Tool Name and arguments manually. Alternatively, check Enable Intelligent Execute and select a model.
  2. Select Next. Next stays disabled until the name is available and the required fields are filled in.
  3. On the Review step, check the type, display name, description, family, connection, and execution mode.
  4. Select Save to Library.

Custom tool

The custom wizard has four steps: Type, Basics, Implementation, and Review.

  1. On Basics, enter the Display name, Name, Description, and Tool family, and then select Next.
  2. On Implementation, select the Tool package and Tool class. The lists come from Python plugin packages that declare tool classes. If no package is listed, deploy a tool package from Platform > Packages (Plugins tab).
  3. Select Next, review the summary, and select Save to Library.
  4. Open the new tool's editor to configure its resource references and settings.

Editing a Tool

  1. On the Tool Library tab, select the pencil icon on the tool card. The tool editor opens at /tools/{toolName}/edit.
  2. Change the settings. The page notes that these are the defaults used when the tool is added to an agent and that existing agents remain unchanged.
  3. Select Save to save and return to the Tools page, or Cancel to discard changes.

The Tool family is read-only for tools that come from an artifact package. For other tools, a family is required; saving without one shows "Select a tool family before saving."

Tool Editor Reference

The same editor is used on the Tools page and in the agent editor. Sections appear based on the tool type.

Basics

Field Description
Display name The name shown in the library. Required.
Name The library resource name. Set only when creating a tool; shown as "Name: …" afterwards.
Description What the tool does. The agent's model reads this to decide when to call the tool. Required.
Tool family The family folder for the tool. Not shown in the agent editor.

Connection (MCP tools)

Under Configuration type, choose one option:

Option Fields
Registered endpoint MCP Server Endpoint (lists endpoints from Agents > MCP Servers) and Transport
Inline URL MCP Server URL, Transport, Connection Timeout (seconds), and HTTP Headers

Transport is Streamable HTTP or HTTP SSE.

Registered endpoints are recommended because the URL, headers, and authentication are managed once on the MCP server resource. An inline URL must be reachable from the Management Portal backend, so localhost and private network addresses do not work unless the portal runs in that network. To move an inline configuration to a registered endpoint, select Convert to registered endpoint, enter an Endpoint name, and select Convert. The endpoint and any header secrets are created when you save the tool or agent; canceling before saving creates nothing.

See MCP Servers for endpoint authentication and HTTP header options.

Tool Selection and Configuration (MCP tools)

Enable Intelligent Execute controls how the MCP tool runs:

  • Unchecked (direct mode). The agent's model calls one MCP tool. Configure:
    • List Tools discovers the server's tools. Pick one from the list and select Configure to fill in the description, MCP Tool Name, Response Format, and Tool Arguments from the tool's input schema.
    • MCP Tool Name: the tool name on the server.
    • Response Format: JSON or Text.
    • Operation Timeout (seconds): how long a call can run.
  • Checked (Intelligent Execute). An LLM discovers the server's tools, plans the calls, and summarizes the results. Under Intelligent Execute Resources, select the main model. The main and router prompts default to Platform-Agent-Tool-MCP-Main and Platform-Agent-Tool-MCP-Router.

Tool inputs (arguments)

In direct mode, Tool Arguments define the inputs the agent's model must supply. Select + Add Argument to add a row, or the × icon to remove one.

Column Description
Name The argument name the MCP tool expects
Description How the model should extract the value from the user's message
Type String, Number, Integer, Boolean, Array, or Object
Required Whether the model must supply a value
Default Optional value used when the user does not provide one

When you save an MCP tool on an agent, the portal builds the tool's input schema from these arguments. Arguments named conversation_id or session_id are left out of the schema because those values are sent as HTTP headers instead.

For other tools, the input schema comes from the tool class. The portal shows it as Arguments (read-only) and refreshes it from the platform when you save the tool on an agent.

Resource References

Non-MCP tools list the resources they need as slots, for example Main Prompt, Router Prompt, and Main Model. Required slots are marked with *.

  • Prompt slots have a searchable list (type 3 or more characters) and View and Edit buttons that open the prompt in a side panel.
  • AI model slots list all AI models and include a Temperature setting.

Built-in tools fill in their platform prompts by default (for example Platform-Agent-Tool-GeminiImageGen-Main and Platform-Agent-Tool-GeminiImageGen-Router). Most require you to choose the Main Model.

Content Safety

Field Description
Shield tool output Runs the tool's output through a shielding plugin before the agent uses it
Shielding plugin The shielding plugin to use. Shown when shielding is on. Select the refresh icon to reload its parameters.
Parameters The selected plugin's parameters

Settings

Tool-specific properties, such as Grounding mode for Gemini Web Grounding or Allowed Tools for Programmable Tool Calling. The Knowledge tool uses dedicated editors for its vector store query, knowledge graph query, and knowledge unit vector store filters.

Claude Agent Configuration

The Claude Agent tool replaces most sections with a tabbed Claude Agent Configuration panel. See Configure the Claude Agent Tool.

Tool Sets

A tool set is a named bundle of library tools. The Tool Sets tab shows a card for each set with its display name, origin, and description.

Creating or editing a tool set

  1. On the Tool Sets tab, select Create set, or select the pencil icon on an existing set.
  2. In the Create Tool Set or Edit Tool Set dialog, enter the Display name. When creating, review the generated Name and its availability.
  3. Optionally enter a Description.
  4. Under Tools in set, pick tools from Add a tool.... Each row has these columns:
Column Description
Tool The library tool
Family The tool's family
Agent tool name The name the tool gets on an agent. Leave blank to use the suggested name shown in the placeholder.
Initially ignored When checked, the tool stays in the set but is left off the agent when the set is added to an agent or used to create a self-service agent. It can still be added later.
  1. Select Save. To delete a set from this dialog, select Delete set.

Deleting a tool set removes the set only. The tools in the library are not affected.

The built-in sets are:

Set Resource name Tools (agent tool name)
Perplexity Essentials Platform-ToolSet-Perplexity-Essentials Perplexity Search (Search), Perplexity Ask (Ask)
Agents Self-Service Platform-ToolSet-SelfService Code (Code), Knowledge (Knowledge), Perplexity Ask (Ask), Perplexity Search (Search), Gemini Image Generation (ImageGen)
Cursor Essentials Platform-ToolSet-Cursor-Essentials All Cursor MCP tools

Adding Tools to an Agent

Tools are attached on the agent editor's Tools tab. The tab shows the Self-service tool set selector, the Add Tool button, and the Tools available to the agent table with Name, Type, and Actions columns.

Add a tool from the library

  1. Open the agent for editing (or create a new one) and select the Tools tab.
  2. Select Add Tool. The Add Tool to Agent dialog opens with the tool library folders and a search box.
  3. Expand a folder and select Select on the tool you want.
  4. Select Add to Agent. The tool is copied to the agent and the Edit Agent Tool dialog opens.
  5. Review the Name (the agent tool name), Description, and the other sections, and fill in required values such as the Main Model.
  6. Select Save changes. The toast "Changes apply to this agent copy only." confirms the update.
  7. Select Save Changes (or Create Agent) on the agent to persist the tools.

Add a tool set

  1. In the Add Tool to Agent dialog, under More options, select Add Tool Set.
  2. Select Add set next to the set. Every tool in the set that is not initially ignored is copied to the agent with the set's agent tool name. A toast reports how many tools were copied and how many were left off.
  3. The editor opens for the last tool added. Configure each tool and save the agent.

Copy tools from another agent

  1. In the Add Tool to Agent dialog, select Copy from another agent.
  2. Select source: choose the Source agent and check the tools to copy.
  3. Review differences: for tools whose name already exists on the agent, choose Use source, Keep target, or Skip. Optionally check Also save copy to Tool Library.
  4. Confirm: select Copy Tools to Agent.

Edit or remove an agent tool

In the Tools available to the agent table, select the pencil icon to open Edit Agent Tool, or the trash icon to remove the tool. Removing a tool from the agent does not change the Tool Library.

Self-service tool set

The Self-service tool set selector links a tool set to the agent. It does not add tools by itself. Self-service flows, such as agents created from an agent template that references a self-service tool set, use this set to decide which tools can be enabled or disabled on the agent. Only tools in the set can be managed that way. Clear the selector to remove the link.

Tool Name Recommendations

Two names matter for every tool:

  • Library resource name (the Name field when creating a tool, family, or set). The portal derives it from the display name in Title-Case with hyphens, for example "Perplexity Search" becomes Perplexity-Search. Use letters, numbers, hyphens, and underscores, with no spaces. Built-in resources use the Platform-Tool-, Platform-ToolFamily-, and Platform-ToolSet- prefixes; use your own prefix for organization resources so they are easy to tell apart.
  • Agent tool name (the Name field in Edit Agent Tool). This is the function name the agent's model calls. It must start with a letter and can contain only letters, numbers, hyphens, and underscores; other characters are removed as you type. It must be unique on the agent.

When you add a tool, the portal suggests an agent tool name (for example GeminiImageGen, ClaudeAgent, or Search) and appends a number such as Search2 when the name is already taken. Keep agent tool names short and distinct, and make each description specific so the model can tell the tools apart.

Reserved name: end_conversation

Do not name an agent tool end_conversation (in any letter case). The platform registers a tool with that name to end realtime voice sessions. The Management Portal does not block the name, but when an agent with realtime speech enabled has a tool called end_conversation, the Core API rejects the voice session with an error asking you to rename the agent tool. See Enabling Real-Time Speech.