Tools
External integrations, MCP servers, and custom tool implementations.
Tools extend your AI pipelines with external capabilities. During an inference step, the LLM can decide to call a tool to look up data, perform actions, or interact with external systems. MechaMental manages tools through a Library and Instance model that separates tool definitions from workspace-specific configurations.
Tool Library vs Your Tools
The Tools page has two tabs:
- Your Tools — tool instances configured in your workspace, shown in a table with name, type, status (Active/Inactive), and usage count.
- Library — the global catalog of available tools, displayed as cards organized by category. Each card shows the tool name, description, version, category badge, and an Install or Add button.
Multiple Instances
You can create multiple instances of the same library tool with different configurations. For example, two Slack instances pointing at different workspaces, or a database tool configured for production vs staging. Tools that already have instances show an instance count badge and an Add button instead of Install.
Tool Categories
The library organizes tools into filterable categories:
| Category | Description |
|---|---|
| AI & ML | Image generation, speech-to-text, embeddings |
| Data | SQL databases, Elasticsearch, data APIs |
| Communication | Slack, Email, SMS, messaging platforms |
| Storage | File storage, object stores, S3 |
| Development | GitHub, Jira, CI/CD triggers |
| Utilities | Data transforms, formatters, general-purpose tools |
Use the category filter buttons above the library grid or the search bar to narrow down results.
Installing a Tool
When you click a tool card in the Library (or click Install / Add), the install dialog opens. The dialog title reflects the context:
- Install [Tool Name] — first time adding this tool
- Add [Tool Name] Instance — when instances already exist
- Edit [Tool Name] — when editing an existing instance
Configure Basic Details
Fill in the instance details:
| Field | Required | Description |
|---|---|---|
| Instance Name | Yes | A name for this specific instance |
| Description | No | What this instance is configured for |
| Namespace | No | Optionally scope the instance to a specific namespace |
Set Configuration Fields
Each tool defines a JSON schema for its configuration. The dialog auto-generates form fields from this schema. Fill in tool-specific settings such as API endpoints, connection strings, or behavioral parameters.
Map Credentials
Tools that require authentication show credential selector fields. These reference secrets stored in the Vault so that sensitive values (API keys, tokens) are never stored in the tool configuration directly. The dialog fetches available active secrets from your workspace vault.
Configure MCP Overrides (Optional)
For MCP-type tools, additional override fields are available:
- MCP Description Override — custom description for the tool in the LLM's context
- MCP Action Filter — limit which MCP actions are exposed
- MCP Action Descriptions — override individual action descriptions
- MCP Max Calls Per Turn — limit how many times the tool can be called per inference turn
- MCP Requires Confirmation — require user confirmation before execution
- MCP Parameter Overrides — override default parameter values
Set Config Profiles
After the instance is created, you can configure environment-specific profiles. Each profile maps to an environment (e.g., development, staging, production) and contains its own configuration values. This lets you use sandbox APIs in development and production APIs in production without changing your pipeline.
Credential Mapping
Tools that require authentication can reference secrets stored in the Vault. Credential mapping connects tool configuration fields to vault secrets. When a tool executes, the platform resolves credential references to their current secret values at runtime.
Using Tools in Pipelines
Tools are used in pipelines through two mechanisms:
| Usage | Description |
|---|---|
| Inference steps | Attach tool instances to an LLM step. The model decides when and how to call them based on the conversation context. |
| Tool call steps | Directly invoke a specific tool with predetermined parameters. The tool runs unconditionally as part of the pipeline flow. |
In the pipeline editor, inference steps have a tools configuration section where you select which tool instances to make available. Tool call steps let you pick a specific tool and action, then map input parameters.