Practical MCP with FastMCP & LangChain

How Do You Get a Context Object?

You add ctx, context, or any name you like to your tool function signature as a parameter with a default value. FastMCP uses dependency injection to provide it automatically:

from fastmcp import FastMCP
from fastmcp.dependencies import CurrentContext
from fastmcp.server.context import Context

mcp = FastMCP("my-server")

@mcp.tool()
async def my_tool(query: str, ctx: Context = CurrentContext()) -> str:
    # ctx is injected by FastMCP — the client never passes it
    await ctx.info(f"Searching for: {query}")
    return f"Results for {query}"

The critical thing to understand: ctx is not a normal parameter. The client never passes it to the tool. It doesn't show up in the tool's schema. FastMCP injects it automatically at runtime, the same way a web framework injects a "request" object. From the client's point of view, my_tool takes one parameter: query.

Because ctx methods like info(), set_state(), and report_progress() are async, you should declare your function with async def when using context so you can await them. That said, FastMCP does run synchronous tools in a thread pool with context propagation, so injecting and accessing Context from a def function is possible — but using async def gives you the most natural access pattern.

This same pattern works with tools, resources, and prompts. You are not limited to using context only inside tools:

from fastmcp import FastMCP
from fastmcp.dependencies