Custom tools

Custom tools extend agent capabilities. You can create tools once and reuse them across multiple agents with consistent behavior and governance.

Lambda tools

User-defined Python functions that execute in secure, sandboxed environments:

• Run in isolated Python 3.12 environments.
• Automatic schema generation from type annotations.
• No network access.
• Resource limits: memory fixed by the execution environment, 30-second timeout (configurable up to 21600 seconds).

Use Lambda tools for custom business logic, data transformations, and calculations that don't require external connectivity.

MCP tools

Tools from external MCP-compliant servers that enable enterprise integrations:

• Standards-based protocol.
• Dynamic tool discovery.
• Register a tool server once, automatically discover all its tools.
• Connect to CRM systems, databases, and enterprise APIs.

Register an MCP tool server to make its tools available. Learn more about MCP and Tool Servers.

Using custom tools with agents

After creating a Lambda tool or registering an MCP server, reference the tool in your agent configuration:

LAMBDA TOOL CONFIGURATION IN AGENT

Code example with json syntax.json

1

MCP TOOL CONFIGURATION IN AGENT

Code example with json syntax.json

1

Argument overrides

Control tool behavior with hardcoded values or dynamic context references:

TOOL CONFIGURATION WITH ARGUMENT OVERRIDE

Code example with json syntax.json

1

Argument overrides:

• Hardcoded values that the LLM cannot modify (e.g., "customer_tier": "enterprise").
• Dynamic references using $ref to pull values from:
  • Session metadata: {"$ref": "session.metadata.field_name"}
  • Agent metadata: {"$ref": "agent.metadata.field_name"}
  • Agent secrets: {"$ref": "agent.secrets.secret_name"} — see Agent secrets for the encrypted, masked storage that backs this.

For complete API details, see Tools API.