Anthropic Claude Tool Use Errors — Fix tool_result, input_schema & Streaming

The multi-turn tool loop structure

When Claude's response has stop_reason: "tool_use", you must: (1) append the full assistant message to your messages array, (2) execute the tool, (3) add a new user message containing a tool_result block. Sending any other user message in this position causes the validation error.

import anthropic, json

client = anthropic.Anthropic()

tools = [{
    "name": "get_weather",
    "description": "Get current weather for a city.",
    "input_schema": {
        "type": "object",
        "properties": {
            "city": {"type": "string", "description": "City name"}
        },
        "required": ["city"]
    }
}]

messages = [{"role": "user", "content": "What is the weather in Paris?"}]

# --- agentic loop ---
while True:
    response = client.messages.create(
        model="claude-opus-4-8",
        max_tokens=1024,
        tools=tools,
        messages=messages,
    )

    # Always append the assistant turn first
    messages.append({"role": "assistant", "content": response.content})

    if response.stop_reason == "end_turn":
        # No more tool calls — print final text and exit
        for block in response.content:
            if block.type == "text":
                print(block.text)
        break

    if response.stop_reason == "tool_use":
        # Collect ALL tool_result blocks for this turn
        tool_results = []
        for block in response.content:
            if block.type == "tool_use":
                # Execute your tool here
                result = {"temperature": "18C", "condition": "Partly cloudy"}
                tool_results.append({
                    "type": "tool_result",
                    "tool_use_id": block.id,
                    "content": json.dumps(result),
                })
        # Send all results back in one user message
        messages.append({"role": "user", "content": tool_results})

tool_choice: auto vs any vs specific tool

The tool_choice parameter controls whether Claude must use a tool:

# auto (default) — Claude decides whether to use a tool
response = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    tools=tools,
    tool_choice={"type": "auto"},
    messages=messages,
)

# any — Claude must use at least one tool (no plain text response)
response = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    tools=tools,
    tool_choice={"type": "any"},
    messages=messages,
)

# tool — force a specific tool by name
response = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    tools=tools,
    tool_choice={"type": "tool", "name": "get_weather"},
    messages=messages,
)

• Use "any" for structured extraction: combine tool_choice: any with a single extraction tool to force Claude to always return structured JSON instead of prose.
• disable_parallel_tool_use: add "disable_parallel_tool_use": true inside tool_choice to guarantee at most one tool per response, simplifying your loop logic.

Parallel tool use — return all results in one user message

Claude may return multiple tool_use blocks in a single response. You must execute all of them and return all results in one user message. Sending partial results causes a validation error on the following call.

import asyncio

async def run_tools_parallel(response_content, tool_executor):
    """Execute all tool_use blocks concurrently, return list of tool_result blocks."""
    tool_use_blocks = [b for b in response_content if b.type == "tool_use"]

    async def execute_one(block):
        result = await tool_executor(block.name, block.input)
        return {
            "type": "tool_result",
            "tool_use_id": block.id,
            "content": json.dumps(result),
        }

    return await asyncio.gather(*[execute_one(b) for b in tool_use_blocks])

# Usage:
# tool_results = await run_tools_parallel(response.content, my_executor)
# messages.append({"role": "user", "content": tool_results})

Computer use — beta header required

The computer use tool (computer_20241022, text_editor_20241022, bash_20241022) requires a beta header. Without it the API returns a 400 error.

response = client.beta.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    betas=["computer-use-2024-10-22"],  # required beta header
    tools=[
        {"type": "computer_20241022", "name": "computer", "display_width_px": 1280, "display_height_px": 800},
        {"type": "text_editor_20241022", "name": "str_replace_editor"},
        {"type": "bash_20241022", "name": "bash"},
    ],
    messages=messages,
)