Skip to main content

adapters.pydantic_ai_adapter

Pydantic AI agent adapter.

Wraps a Pydantic AI Agent to implement the BaseAgent interface, enabling use with protocol adapters.

PydanticAIAdapter Objects​

class PydanticAIAdapter(BaseAgent)

Adapter for Pydantic AI agents.

Wraps a pydantic_ai.Agent to provide a consistent interface for protocol adapters.

Example:

from pydantic_ai import Agent

pydantic_agent = Agent("openai:gpt-4o", instructions="...")

agent = PydanticAIAgent( agent=pydantic_agent, name="my_agent", description="A helpful assistant", )

response = await agent.run("Hello!", context)

__init__​

def __init__(agent: Agent,
             name: str = "pydantic_ai_agent",
             description: str = "A Pydantic AI powered agent",
             version: str = "1.0.0",
             agent_id: str | None = None,
             selected_mcp_servers: list[Any] | None = None,
             non_mcp_toolsets: list[Any] | None = None,
             codemode_builder: Any = None)

Initialize the Pydantic AI agent adapter.

Arguments:

  • agent - The Pydantic AI Agent instance.
  • name - Agent name.
  • description - Agent description.
  • version - Agent version.
  • agent_id - Unique identifier for usage tracking (defaults to name).
  • selected_mcp_servers - List of MCP server selections to use.
  • non_mcp_toolsets - List of non-MCP toolsets (e.g., codemode, skills) to always include.
  • codemode_builder - Optional callable to rebuild codemode toolset when MCP servers change.

name​

@property
def name() -> str

Get the agent's name.

description​

@property
def description() -> str

Get the agent's description.

version​

@property
def version() -> str

Get the agent's version.

agent_id​

@property
def agent_id() -> str

Get the agent's unique identifier for usage tracking.

selected_mcp_server_ids​

@property
def selected_mcp_server_ids() -> list[str]

Get the list of selected MCP server IDs.

codemode_enabled​

@property
def codemode_enabled() -> bool

Check if codemode is currently enabled.

set_codemode_enabled​

def set_codemode_enabled(enabled: bool) -> bool

Enable or disable codemode at runtime.

When enabling, builds a new CodemodeToolset using the codemode_builder. When disabling, removes the existing CodemodeToolset.

Arguments:

  • enabled - Whether to enable (True) or disable (False) codemode.

Returns:

True if the operation succeeded, False otherwise.

update_mcp_servers​

def update_mcp_servers(servers: list[Any]) -> None

Update the list of selected MCP servers.

Also rebuilds Codemode toolset if a builder was provided, to ensure tool bindings reflect the current MCP server selection.

Arguments:

  • servers - New list of MCP server selections to use.

pydantic_agent​

@property
def pydantic_agent() -> Agent

Get the underlying Pydantic AI agent.

get_tools​

def get_tools() -> list[ToolDefinition]

Get the list of tools available to this agent.

run​

async def run(prompt: str, context: AgentContext) -> AgentResponse

Run the agent with a prompt.

Arguments:

  • prompt - User prompt/message.
  • context - Execution context.

Returns:

Complete agent response.

stream​

async def stream(prompt: str,
                 context: AgentContext) -> AsyncIterator[StreamEvent]

Run the agent with streaming output.

Uses agent.run() with an event_stream_handler so that the full agent graph executes (multiple model→tool→model rounds) while text deltas are still yielded token-by-token.

run_stream() is intentionally not used here because it stops the graph at the first output matching the output type, which prevents multi-round tool usage.

Arguments:

  • prompt - User prompt/message.
  • context - Execution context.

Yields:

Stream events as they are produced.

run_with_codemode​

async def run_with_codemode(prompt: str, context: AgentContext,
                            codemode_executor: Any) -> AgentResponse

Run the agent with CodeMode for programmatic tool composition.

This variant allows the agent to use agent-codemode for executing code that composes multiple tools efficiently.

Arguments:

  • prompt - User prompt/message.
  • context - Execution context.
  • codemode_executor - CodeModeExecutor instance.

Returns:

Complete agent response.