MechaMental
Guides

Working with Tools

Browse the tool library, install instances, configure credentials, and use tools in pipelines.

Overview

Tools extend your AI pipelines with external capabilities -- APIs, databases, web services, and more. MechaMental uses the MCP (Model Context Protocol) standard for tool integration. This guide covers browsing the library, installing a tool instance, configuring it, and using it in your pipeline.

Tool Page Layout

Navigate to Tools in the sidebar. The page has two tabs:

TabWhat It Shows
Your ToolsTool instances you have installed and configured in this workspace
LibraryAll available tools you can browse and install

You can also click Browse Library in the top-right header to jump directly to the Library tab.

Installing a Tool Instance

Browse the Library

  1. Navigate to Tools in the sidebar
  2. Switch to the Library tab
  3. Browse or search by name, category, or description
  4. Each tool card shows the tool name, description, category badge, and available actions

Open the Install Dialog

Click the Install button on the tool you want to add. The install dialog opens, pre-filled with information from the library tool.

Configure the Instance

Fill in the instance configuration fields:

  • Instance Name -- a descriptive name for this specific installation (e.g., "Production Slack", "Dev Database")
  • Description -- what this instance is used for
  • Namespace -- optional scope for the tool instance
  • Config -- tool-specific configuration fields (URLs, parameters, etc.)
  • Credentials -- sensitive fields like API keys and tokens

Credential Security

Credentials are stored securely and are never exposed in the UI after saving. Map them to Vault secrets for environment-specific rotation.

Confirm Installation

Click Install to save the instance. It appears in the Your Tools tab as an active instance, ready to be used in pipelines.

Using Tools in an Inference Step

The most common way to use tools is by attaching them to an inference step so the LLM can decide when to call them.

Open the Step Detail Dialog

In the pipeline editor, click on an inference step (or create a new one) to open the step detail dialog.

Select Tools

  1. Switch to the Tools tab (labeled Tools & Skills)
  2. The tab lists all active tool instances in your workspace
  3. Use the search bar (placeholder: "Search tool instances...") to filter
  4. Click a tool row to select it -- a checkbox and amber highlight indicate selection
  5. Selected tools appear as amber badge chips at the top with an "X" to remove

Configure Execution Settings

When at least one tool is selected, the Execution Settings section appears:

  • Tool Execution Mode:
    • Single Turn -- the LLM calls a tool once, gets the result, then responds
    • Multi Turn (Agentic) -- the LLM can call tools in a loop until it decides it has enough information
  • Max Tool Rounds (multi-turn only): the maximum number of rounds before forcing a final answer (1--10, default 3)

Using Tools in a Tool Call Step

For deterministic tool invocations where you always want to call a specific tool (regardless of LLM judgment), use a Tool Call step:

  1. Add a new step to your pipeline and select Tool Call as the step type
  2. In the Config tab, select the tool instance
  3. Define the Input Mapping using a Jinja template to build the tool's input from the execution context

MCP Overrides

When attaching tools to inference steps, you can apply per-step overrides:

  • Description Override -- replace the tool's default description with context-specific instructions for the LLM
  • Action Filter -- restrict which tool actions are available (useful for tools with many capabilities)
  • Max Calls Per Turn -- limit how many times this specific tool can be called in a single turn

Credential Management

Instead of hardcoding API keys in tool configurations, use Vault secrets:

  1. Store your API key in the workspace Vault (Settings area)
  2. In the tool instance configuration, reference the vault secret
  3. Different environments can map to different vault secrets (e.g., sandbox API key for dev, production key for prod)

Next Steps

Managing Sources

Upload files, paste text, configure chunking and indexing, and monitor processing status.

Deploying a Release

Create an immutable release, deploy to an environment, test, and roll back if needed.

On this page

OverviewTool Page LayoutInstalling a Tool InstanceBrowse the LibraryOpen the Install DialogConfigure the InstanceConfirm InstallationUsing Tools in an Inference StepOpen the Step Detail DialogSelect ToolsConfigure Execution SettingsUsing Tools in a Tool Call StepMCP OverridesCredential ManagementNext Steps