Overview

Tools give an agent the ability to interact with the outside world beyond reasoning: web search, code execution, image and video generation, access to third-party services, and more. VeADK packages these capabilities as tools you can register directly. At runtime the agent decides on its own — based on user input and its instruction — whether to call a tool; the framework executes it automatically and feeds the result back into the reasoning process.

Adding tools to an agent

Whatever the tool type, registration is the same: put the tool in the Agent’s tools list.

from veadk import Agent
from veadk.tools.builtin_tools.web_search import web_search

agent = Agent(tools=[web_search])

Once registered, the agent decides when to call a tool and how to pass arguments, based on the function signature / tool description, the user input, and its instruction. Multiple tools can be registered at once, and the agent picks among them as needed during reasoning.

Tool categories

VeADK tools are grouped by how they connect and what they do.

Category	Description	Docs
Built-in tools	Pre-packaged function tools for web search, image and video generation, code sandboxes, and more	Web search
Built-in MCP tools	MCP toolsets for Feishu, LAS, VOD, MCP Router, Supabase, and more	Feishu tools
Custom tools	Wrap any Python function as a tool to extend the agent	Custom tools
Dynamic agents	A built-in toolset that lets a main agent collect resources, create sub-agents, and delegate a task at runtime	Dynamic agent delegation
System tools	Knowledge-base and long-term-memory retrieval tools mounted automatically from agent configuration	Load a knowledge base
Content safety	Callback-based checks that block unsafe input and output	Content safety
Feishu Channel extension	Bridge inbound Feishu bot messages into the `Runner`	Feishu Channel extension

Built-in tools and built-in MCP tools mostly depend on Volcengine services — enable the corresponding service and configure authentication (AK/SK, API keys) before use. System tools are managed by the framework and need not be added to the tools list manually.

Preparation and result checks

Complete installation and model configuration, then enable the service and credentials listed on the tool page. Model API keys and tool AK/SK or service API keys are separate. Setting CLOUD_PROVIDER does not switch every tool endpoint automatically. Agent(tools=[...]) registers capabilities without executing them. After Runner.run(), inspect execution events or tool results to confirm the call and its status. A model answer alone does not prove an external operation succeeded. Await asynchronous tools, and call await toolset.close() when shutting down an MCP application. Start with the custom calculator to check the calling flow. For MCP, discover tools before invoking a capability with the required permissions. Sandbox, mobile, and third-party service tools can change real files or account data; specify the allowed operations in the tool instructions.

Get started

Core components

Observability

Extensions

Frontend

Security

Deployment

Overview

Adding tools to an agent

Tool categories

Preparation and result checks

​Adding tools to an agent

​Tool categories

​Preparation and result checks

Adding tools to an agent

Tool categories

Preparation and result checks