runtime

Runtime context passed to graph builder factories within the Agent Server.

Requires version 0.7.30 or later of the agent server.

The server calls your graph factory in multiple contexts: executing runs, reading state, fetching schemas, and more. ServerRuntime provides the authenticated user, store, and access context for every call. Use .execution_runtime to narrow to the execution variant and access context.

Example — conditionally initialize MCP tools only during execution:

import contextlib
from dataclasses import dataclass

from langchain.agents import create_agent
from langgraph_sdk.runtime import ServerRuntime
from my_agent import connect_mcp, disconnect_mcp

@dataclass
class MyCtx:
    mcp_endpoint: str

_readonly_agent = create_agent("anthropic:claude-3-5-haiku", tools=[])

@contextlib.asynccontextmanager
async def my_factory(runtime: ServerRuntime[MyCtx]):
    if ert := runtime.execution_runtime:
        # Only connect to MCP servers for actual runs.
        # Schema / graph introspection calls skip this.
        user_id = runtime.ensure_user().identity
        mcp_tools = await connect_mcp(ert.context.mcp_endpoint, user_id)
        yield create_agent("anthropic:claude-3-5-haiku", tools=mcp_tools)
        await disconnect_mcp()
    else:
        yield _readonly_agent

Example — simple factory that ignores context:

from langgraph_sdk.runtime import ServerRuntime

def build_graph(user: BaseUser) -> CompiledGraph:
    ...

async def my_factory(runtime: ServerRuntime) -> CompiledGraph:
    # No generic needed if you don't use context.
    return build_graph(runtime.ensure_user())

The base ASGI user protocol