Custom tools and agent skills

The SDK lets you extend your agent’s capabilities using custom Python functions, built-in web tools, built-in system tools, and agent skills.

For Model Context Protocol (MCP) server configuration, see the MCP Documentation.

Built-in tools reference

The SDK provides identifiers for built-in system tools through the BuiltinTools enum (google.antigravity.types.BuiltinTools).

Identifier	Enum constant	Description
`list_directory`	`BuiltinTools.LIST_DIR`	List directory contents.
`search_directory`	`BuiltinTools.SEARCH_DIR`	Search within files.
`find_file`	`BuiltinTools.FIND_FILE`	Find files by pattern.
`view_file`	`BuiltinTools.VIEW_FILE`	Read file contents.
`create_file`	`BuiltinTools.CREATE_FILE`	Create a new file.
`edit_file`	`BuiltinTools.EDIT_FILE`	Edit an existing file.
`run_command`	`BuiltinTools.RUN_COMMAND`	Execute a shell command.
`ask_question`	`BuiltinTools.ASK_QUESTION`	Prompt user for input.
`start_subagent`	`BuiltinTools.START_SUBAGENT`	Invoke a child subagent.
`generate_image`	`BuiltinTools.GENERATE_IMAGE`	Generate or edit images.
`search_web`	`BuiltinTools.SEARCH_WEB`	Perform Google Search.
`read_url_content`	`BuiltinTools.READ_URL_CONTENT`	Fetch URL content.
`finish`	`BuiltinTools.FINISH`	Return final output.

Tool group helpers and filtering

In some cases you may want to restrict the tools that your agent can access. You can control the agent’s active tools in CapabilitiesConfig with enabled_tools or disabled_tools.

For example, you can configure your agent to only have access to read_only built-in tools:

from google.antigravity import LocalAgentConfig, CapabilitiesConfig
from google.antigravity.types import BuiltinTools

config = LocalAgentConfig(
    capabilities=CapabilitiesConfig(
        enabled_tools=BuiltinTools.read_only()
    )
)

Custom Python functions

In some cases you may want to create custom Python functions that your agent can leverage. You can register custom Python functions in LocalAgentConfig as tools for your agent.

For example, you can configure your agent to leverage a custom get_weather function:

from google.antigravity import Agent, LocalAgentConfig

def get_weather(city: str) -> str:
    """Returns the current weather for a city."""
    return f"It's sunny in {city}."

config = LocalAgentConfig(tools=[get_weather])

async with Agent(config) as agent:
    response = await agent.chat("What's the weather in Tokyo?")
    print(await response.text())

Built-in web tools

Web search (SEARCH_WEB) and URL fetching (READ_URL_CONTENT) tools are enabled by default.

For example, you can ask your agent to perform Google searches or read web pages:

from google.antigravity import Agent, LocalAgentConfig

config = LocalAgentConfig()

async with Agent(config) as agent:
    response = await agent.chat("Search for latest updates on Gemini.")
    print(await response.text())

Agent skills

In LocalAgentConfig, skills_paths accepts paths to individual skill directories containing a SKILL.md file, as well as parent directories containing multiple skill subdirectories.

For example, you can configure your agent to leverage a custom code-review skill:

from google.antigravity import Agent, LocalAgentConfig

config = LocalAgentConfig(
    skills_paths=["/path/to/skills/code-review"],
)

async with Agent(config) as agent:
    response = await agent.chat("Run a code quality audit.")
    print(await response.text())

Sample code

For full working code examples, see the GitHub repository: