Tool Call Rendering

Render your agent's tool calls with custom UI components.

Tool Rendering

Waking up the demo

What is this?

Tools are how an LLM invokes predefined, typically-deterministic functions. Tool rendering lets you decide how each of those tool calls appears in the chat. By default, a tool call does not show in the chat. You register a React component that renders a branded card for the call (arguments, live status, and the eventual result). This is the Generative UI variant CopilotKit calls tool rendering.

When should I use this?

Render tool calls when you want to:

  • Show users exactly what tools the agent is invoking and with what arguments
  • Display live progress indicators while a tool executes
  • Render rich, polished results once a tool completes
  • Give tool-heavy agents a transparent, on-brand chat experience

Default tool rendering (zero-config)

The simplest entry point: call useDefaultRenderTool() with no arguments. CopilotKit registers its built-in DefaultToolCallRenderer as the * wildcard: every tool call renders as a tidy status card (tool name, live Running → Done pill, collapsible arguments/result) without you writing any UI. Calls the agent makes in a row are grouped into one card; see Grouping tool calls.

Without this hook, the chat has no * renderer, and tool calls without a named renderer are invisible. The user only sees the assistant's final text summary.

page.tsx
  // Opt in to CopilotKit's built-in default tool-call card. Called with
  // no config so the package-provided `DefaultToolCallRenderer` is used
  // as the wildcard renderer — this is the "out-of-the-box" UI the cell
  // is meant to showcase.
  useDefaultRenderTool();

Here's what the built-in status card looks like for each tool call:

Tool Rendering (Default Catch-all)

Waking up the demo

Custom catch-all

Once you want on-brand chrome, pass a render function to useDefaultRenderTool. It's a convenience wrapper around useRenderTool({ name: "*", ... }): one wildcard renderer handles every tool call, named or not:

page.tsx
  // `useDefaultRenderTool` is a convenience wrapper around
  // `useRenderTool({ name: "*", ... })` — a single wildcard renderer
  // that handles every tool call not claimed by a named renderer.
  useDefaultRenderTool(
    {
      render: ({ name, parameters, status, result }) => (
        <CustomCatchallRenderer
          name={name}
          parameters={parameters}
          status={status as CatchallToolStatus}
          result={result}
        />
      ),
    },
    [],
  );

The catch-all component reads the tool name, status, parameters, and result. It uses the shadcn/ui Card and Badge components:

custom-catchall-renderer.tsx
"use client";

import React from "react";
import {
  Card,
  CardContent,
  CardDescription,
  CardHeader,
  CardTitle,
} from "./_components/card";
import { Badge } from "./_components/badge";

// ShadCN-styled catch-all renderer for the tool-rendering-custom-catchall
// cell. A single wildcard renderer handles every tool call — name,
// status, arguments, and result rendered inside a shadcn <Card />.

export type CatchallToolStatus = "inProgress" | "executing" | "complete";

export interface CustomCatchallRendererProps {
  name: string;
  status: CatchallToolStatus;
  parameters: unknown;
  result: string | undefined;
}

export function CustomCatchallRenderer({
  name,
  status,
  parameters,
  result,
}: CustomCatchallRendererProps) {
  const parsedResult = parseResult(result);
  const done = status === "complete";

  return (
    <Card
      data-testid="custom-wildcard-card"
      data-tool-name={name}
      data-status={status}
      className="my-3 overflow-hidden"
    >
      <CardHeader className="flex flex-row items-center justify-between space-y-0 border-b border-neutral-200 dark:border-border bg-neutral-50/60 dark:bg-muted/60 py-3">
        <div className="flex items-center gap-2">
          <CardTitle
            data-testid="custom-wildcard-tool-name"
            className="font-mono text-sm text-neutral-900 dark:text-foreground"
          >
            {name}
          </CardTitle>
          <CardDescription className="text-[10px] uppercase tracking-wider text-neutral-500 dark:text-muted-foreground">
            tool call
          </CardDescription>
        </div>
        <StatusBadge status={status} />
      </CardHeader>

      <CardContent className="grid gap-3 p-4 text-sm">
        <Section label="Arguments">
          <pre
            data-testid="custom-wildcard-args"
            className="overflow-x-auto rounded-md border border-neutral-200 dark:border-border bg-neutral-50 dark:bg-muted p-2.5 font-mono text-xs text-neutral-900 dark:text-foreground"
          >
            {safeStringify(parameters)}
          </pre>
        </Section>

        <Section label="Result">
          {done ? (
            <pre
              data-testid="custom-wildcard-result"
              className="overflow-x-auto rounded-md border border-emerald-200 dark:border-emerald-800 bg-emerald-50 dark:bg-emerald-950/40 p-2.5 font-mono text-xs text-neutral-900 dark:text-foreground"
            >
              {parsedResult !== undefined
                ? safeStringify(parsedResult)
                : "(empty)"}
            </pre>
          ) : (
            <p className="text-xs italic text-neutral-500 dark:text-muted-foreground">
              waiting for tool to finish…
            </p>
          )}
        </Section>
      </CardContent>
    </Card>
  );
}

function Section({
  label,
  children,
}: {
  label: string;
  children: React.ReactNode;
}) {
  return (
    <div>
      <div className="mb-1.5 text-[10px] font-medium uppercase tracking-wider text-neutral-500 dark:text-muted-foreground">
        {label}
      </div>
      {children}
    </div>
  );
}

function StatusBadge({ status }: { status: CatchallToolStatus }) {
  const { label, variant, dot } = describeStatus(status);
  return (
    <Badge data-testid="custom-wildcard-status" variant={variant}>
      <span
        className={`inline-block h-1.5 w-1.5 rounded-full ${dot}`}
        aria-hidden
      />
      {label}
    </Badge>
  );
}

function describeStatus(status: CatchallToolStatus): {
  label: string;
  variant: "warning" | "secondary" | "success";
  dot: string;
} {
  switch (status) {
    case "inProgress":
      return {
        label: "streaming",
        variant: "warning",
        dot: "bg-amber-500 animate-pulse",
      };
    case "executing":
      return {
        label: "running",
        variant: "secondary",
        dot: "bg-neutral-500 animate-pulse",
      };
    case "complete":
      return {
        label: "done",
        variant: "success",
        dot: "bg-emerald-500",
      };
  }
}

function parseResult(result: string | undefined): unknown {
  if (result === undefined || result === null) return undefined;
  if (typeof result !== "string") return result;
  try {
    return JSON.parse(result);
  } catch {
    return result;
  }
}

function safeStringify(value: unknown): string {
  try {
    return JSON.stringify(value, null, 2);
  } catch {
    return String(value);
  }
}

Here's the branded catch-all in action, where every tool call gets the same on-brand card:

Tool Rendering (Custom Catch-all)

Waking up the demo

Per-tool renderers

The most expressive path is one renderer per tool name. The demo registers four per-tool renderers: get_weather, search_flights, get_stock_price, and roll_d20. Each renderer receives the tool's parsed arguments, a live status, and (once the agent returns) the result.

The frontend pattern is the same for every backend. This shared, docs-only example includes every component, type, and helper that its renderers use:

components/weather-card.tsx
export interface WeatherCardProps {
  loading: boolean;
  location: string;
  temperature?: number;
  humidity?: number;
  windSpeed?: number;
  conditions?: string;
}

export function WeatherCard({
  loading,
  location,
  temperature,
  humidity,
  windSpeed,
  conditions,
}: WeatherCardProps) {
  return (
    <article className="rounded-xl border p-4">
      <h3 className="font-semibold">{location || "Weather"}</h3>
      {loading ? (
        <p>Fetching weather...</p>
      ) : (
        <dl>
          <div>
            <dt>Conditions</dt>
            <dd>{conditions ?? "--"}</dd>
          </div>
          <div>
            <dt>Temperature</dt>
            <dd>{temperature ?? "--"}&deg;F</dd>
          </div>
          <div>
            <dt>Humidity</dt>
            <dd>{humidity ?? "--"}%</dd>
          </div>
          <div>
            <dt>Wind</dt>
            <dd>{windSpeed ?? "--"} mph</dd>
          </div>
        </dl>
      )}
    </article>
  );
}
components/flight-list-card.tsx
export interface Flight {
  airline?: string;
  flight?: string;
  depart?: string;
  arrive?: string;
  price_usd?: number;
}

export interface FlightListCardProps {
  loading: boolean;
  origin: string;
  destination: string;
  flights: Flight[];
}

export function FlightListCard({
  loading,
  origin,
  destination,
  flights,
}: FlightListCardProps) {
  return (
    <article className="rounded-xl border p-4">
      <h3 className="font-semibold">
        {origin || "?"} → {destination || "?"}
      </h3>
      {loading ? (
        <p>Searching...</p>
      ) : (
        <ul>
          {flights.map((flight, index) => (
            <li key={`${flight.flight ?? "flight"}-${index}`}>
              {flight.airline ?? "--"} {flight.flight ?? ""}:{" "}
              {flight.depart ?? "?"} → {flight.arrive ?? "?"}
              {flight.price_usd !== undefined ? ` ($${flight.price_usd})` : ""}
            </li>
          ))}
        </ul>
      )}
    </article>
  );
}
lib/parse-json-result.ts
export function parseJsonResult<T>(result: unknown): T {
  if (!result) return {} as T;

  try {
    return (typeof result === "string" ? JSON.parse(result) : result) as T;
  } catch {
    return {} as T;
  }
}
app/tool-renderers.tsx
"use client";

import { useRenderTool } from "@copilotkit/react-core/v2";
import { z } from "zod";
import { WeatherCard } from "../components/weather-card";
import { FlightListCard, type Flight } from "../components/flight-list-card";
import { parseJsonResult } from "../lib/parse-json-result";

interface WeatherResult {
  city?: string;
  temperature?: number;
  humidity?: number;
  wind_speed?: number;
  conditions?: string;
}

interface FlightSearchResult {
  origin?: string;
  destination?: string;
  flights?: Flight[];
}

export function ToolRenderers() {
  useRenderTool(
    {
      name: "get_weather",
      parameters: z.object({ location: z.string() }),
      render: ({ parameters, result, status }) => {
        const parsed = parseJsonResult<WeatherResult>(result);
        return (
          <WeatherCard
            loading={status !== "complete"}
            location={parameters?.location ?? parsed.city ?? ""}
            temperature={parsed.temperature}
            humidity={parsed.humidity}
            windSpeed={parsed.wind_speed}
            conditions={parsed.conditions}
          />
        );
      },
    },
    [],
  );

  useRenderTool(
    {
      name: "search_flights",
      parameters: z.object({
        origin: z.string(),
        destination: z.string(),
      }),
      render: ({ parameters, result, status }) => {
        const parsed = parseJsonResult<FlightSearchResult>(result);
        return (
          <FlightListCard
            loading={status !== "complete"}
            origin={parameters?.origin ?? parsed.origin ?? ""}
            destination={parameters?.destination ?? parsed.destination ?? ""}
            flights={parsed.flights ?? []}
          />
        );
      },
    },
    [],
  );

  return null;
}

Mount the renderers anywhere beneath the same CopilotKit provider as your chat. The renderer component returns no layout of its own; it registers the two named renderers for tool calls in the chat:

app/page.tsx
"use client";

import { CopilotChat, CopilotKit } from "@copilotkit/react-core/v2";
import { ToolRenderers } from "./tool-renderers";

export default function Page() {
  return (
    // Use the agent name you registered on CopilotRuntime
    <CopilotKit runtimeUrl="/api/copilotkit" agent="default">
      <ToolRenderers />
      <CopilotChat agentId="default" />
    </CopilotKit>
  );
}

The name you pass to useRenderTool must match the tool name the agent exposes. That is how the chat matches the call to your component.

Per-tool renderers compose with a catch-all: named renderers claim the "interesting" tools and a wildcard handles everything else. In the demo, the catch-all component from Custom catch-all handles every tool without a named renderer:

page.tsx
  // Wildcard catch-all for anything that doesn't match a per-tool
  // renderer above.
  useDefaultRenderTool(
    {
      render: ({ name, parameters, status, result }) => (
        <CustomCatchallRenderer
          name={name}
          parameters={parameters}
          status={status as CatchallToolStatus}
          result={result}
        />
      ),
    },
    [],
  );

Tool inputs and results are separate

In useRenderTool, parameters contains the inputs the agent sent to the tool. It does not change into the tool's return value when status becomes "complete". The completed output arrives separately as result, a string. For a tool that returns JSON, parse that string before reading its fields.

For example, get_weather might receive { "location": "Paris" } and return { "temperature": 22 }. Read parameters.location for the requested city and the parsed result.temperature for the temperature. Adding temperature to the renderer's parameters schema does not copy it from the result.

If a card says complete but still shows placeholder details, check whether those details come from parameters instead of the parsed result. Completion reports that the tool returned; it does not fill the card's props for you.

The backend tool definition

The frontend renderer only sees what the agent sends down. Here's the matching backend definition for get_weather: expose a tool named get_weather, return structured data, and let the frontend renderer with the same name render the card.

weather_tool.py
from typing import Any


# Anthropic tool schema — passed via the `tools` parameter on
# `client.messages.create(...)` / `.stream(...)`. Claude calls this
# tool by name; the runtime dispatches to the matching handler below.
GET_WEATHER_TOOL: dict[str, Any] = {
    "name": "get_weather",
    "description": (
        "Get the current weather for a given location. Useful on its "
        "own for weather questions, and a great companion to "
        "`search_flights` — always consider checking the weather at a "
        "destination the user is flying to."
    ),
    "input_schema": {
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "The city or region to get weather for.",
            },
        },
        "required": ["location"],
    },
}


def get_weather(location: str) -> dict[str, Any]:
    """Handler invoked when Claude calls the `get_weather` tool."""
    return {
        "city": location,
        "temperature": 68,
        "humidity": 55,
        "wind_speed": 10,
        "conditions": "Sunny",
    }

The agent must also register this schema as an executable backend tool. This setup shows the connection.

Register the backend tools

run_with_claude_agent_sdk receives the tool schemas from the agent. It converts each schema to an SDK tool and creates an in-process MCP server. The adapter receives this server through mcp_servers. The matching allowed_tools entries give Claude permission to call the tools.

claude_agent_sdk_adapter.py
async def run_with_claude_agent_sdk(
    input_data: RunAgentInput,
    *,
    system_prompt: str,
    tools: list[dict[str, Any]],
    state: Any,
    model: str,
    execute_tool: ExecuteTool,
    max_turns: int = 10,
) -> AsyncIterator[str]:
    """Run through the official AG-UI Claude adapter and emit SSE chunks."""

    encoder = EventEncoder()
    state_box = {"state": state}
    pending_state_snapshots: list[Any] = []
    sdk_tools = _build_sdk_tools(
        tools,
        execute_tool=execute_tool,
        get_state=lambda: state_box["state"],
        set_state=lambda next_state: _set_state(
            next_state,
            state_box,
            pending_state_snapshots,
        ),
    )

    options: dict[str, Any] = {
        "model": _normalize_claude_agent_sdk_model(model),
        "system_prompt": system_prompt,
        "tools": [],
        "permission_mode": "dontAsk",
        "max_turns": max_turns,
    }

    if sdk_tools:
        options["mcp_servers"] = {
            COPILOTKIT_MCP_SERVER_NAME: create_sdk_mcp_server(
                COPILOTKIT_MCP_SERVER_NAME,
                "1.0.0",
                tools=sdk_tools,
            )
        }
        options["allowed_tools"] = [
            f"{COPILOTKIT_TOOL_PREFIX}{schema['name']}" for schema in tools
        ]

    adapter = ClaudeAgentAdapter(
        name="claude-sdk-python",
        options=options,
    )
    run_input = _with_initial_state(input_data, state)

    async for event in adapter.run(run_input):
        if event.type == EventType.TOOL_CALL_RESULT and pending_state_snapshots:
            yield encoder.encode(
                StateSnapshotEvent(
                    type=EventType.STATE_SNAPSHOT,
                    snapshot=pending_state_snapshots.pop(0),
                )
            )
        yield encoder.encode(event)

Call the CopilotKit tool handler

Each sdk_tool_handler calls the executable CopilotKit handler for its schema. It returns the handler result to Claude as MCP tool content.

claude_agent_sdk_adapter.py
def _make_sdk_tool(
    schema: dict[str, Any],
    *,
    execute_tool: ExecuteTool,
    get_state: Callable[[], Any],
    set_state: Callable[[Any], None],
) -> Any:
    name = schema["name"]
    description = schema.get("description", "")
    input_schema = schema.get("input_schema", {"type": "object", "properties": {}})

    @sdk_tool(name, description, input_schema)
    async def sdk_tool_handler(args: dict[str, Any]):
        try:
            # Offload to a worker thread: execute_tool may run a synchronous
            # anthropic.Anthropic() LLM round-trip (the generate_a2ui branch),
            # which would otherwise block the uvicorn event loop and wedge the
            # :8000 /health endpoint under load.
            result_text, next_state = await asyncio.to_thread(
                execute_tool,
                name,
                dict(args or {}),
                get_state(),
                None,
            )
            if next_state is not None:
                set_state(next_state)
            return {"content": [{"type": "text", "text": result_text}]}
        except Exception as exc:
            return {
                "content": [{"type": "text", "text": str(exc)}],
                "is_error": True,
            }

    return sdk_tool_handler

This MCP path handles compatible requests that use backend tools only. Requests with frontend tools or structured user content use the direct Anthropic path. Requests with aimock transport also use this fallback.

To check it, ask the agent for the weather in San Francisco. The WeatherCard renders with the temperature from result.

Grouping tool calls

Agents often call several tools in a row. By default, CopilotKit collapses each series of calls rendered by the built-in card from useDefaultRenderTool() into one card that reads "Used 3 tools". While a call runs, the card names it ("Running read_file src/app.ts"), and opening it shows a row per call. Calls rendered by your own renderers always show on their own and split the group around them, unless the renderer opts in with group: true:

page.tsx
useDefaultRenderTool({ render: MyToolRow, group: true });

Pick how calls show with toolCallDisplay on the message view:

page.tsx
// "grouped" (default) | "line" | "stacked" | "turn" | "separate"
<CopilotChat messageView={{ toolCallDisplay: "line" }} />
DisplayWhat the user sees
"grouped"One card per series of calls, collapsed to "Used 3 tools". Opens to a row per call.
"line"The same grouping as a quiet line of text, like "Thought for 4 seconds".
"stacked"The latest call on top of a stack of the earlier ones. Opens into a card per call.
"turn"The agent works in the open, then everything before the reply folds into "Worked for 9s".
"separate"Every call gets its own card.

To compare the displays side by side, run the Tool call display stories in the React Storybook. They play a scripted run through the real message view, with a control to switch toolCallDisplay.

To restyle or replace the group, use the group slot of the tool-calls view (it applies to the grouped, line and stacked displays):

page.tsx
// ToolTimeline: your component, given the group's `calls` and a `renderCall` function
<CopilotChat
  messageView={{
    assistantMessage: { toolCallsView: { group: ToolTimeline } },
  }}
/>

See CopilotChatToolCallGroup for its props and an example replacement.

Free course: See this pattern built end-to-end in Build Interactive Agents with Generative UI. This free DeepLearning.AI short course is taught by CopilotKit's CEO. It covers the full Generative UI spectrum (Controlled, Declarative, and Open-Ended).

How is this guide?