langchain-ai/langchain · error · OutputParserException

Function arguments: are not valid JSON. Received…

Error message

Function {raw_tool_call['function']['name']} arguments:

{arguments}

are not valid JSON. Received JSONDecodeError {e}

What it means

Raised by parse_tool_call in openai_tools.py when the raw tool call's 'arguments' string fails json.loads (non-partial mode). The message includes the function name, the raw arguments text, and the underlying JSONDecodeError so you can see exactly what the model emitted that was invalid JSON.

Solutions

1. Raise max_tokens so tool call arguments are not cut off
2. Use strict=False in the parser (or parse_partial_json-based streaming) to tolerate non-JSON-compliant strings
3. Catch OutputParserException, log the raw arguments, and retry the model call or repair the JSON

Example fix

# before
tool_calls = parse_tool_call(raw_tool_call, strict=True)

# after
tool_calls = parse_tool_call(raw_tool_call, strict=False)  # tolerate control chars/newlines

import json
args = raw_tool_call["function"]["arguments"]
if args:
    try:
        json.loads(args)
    except json.JSONDecodeError as e:
        ...  # repair or drop before calling parse_tool_call

from langchain_core.exceptions import OutputParserException
try:
    tc = parse_tool_call(raw_tool_call, strict=False)
except OutputParserException as e:
    invalid = make_invalid_tool_call(raw_tool_call, str(e))  # keep going with an InvalidToolCall

When it happens

Trigger: Model returns tool_call arguments that are malformed JSON (truncated by max_tokens, unquoted keys, Python-repr dicts); arguments contain raw newlines/control characters while strict=True. Note: None or empty-string arguments are treated as {} and do NOT raise; the partial=True path swallows errors instead.

Common situations: Tight max_tokens truncating tool arguments mid-JSON; smaller/local models emitting invalid JSON; strict json.loads rejecting otherwise-usable output with control characters.

Related errors

• {exceptions joined with '\n\n'}
• Could not parse function call data
• This output parser can only be used with a chat generation.
• Tool arguments must be specified as a dict, received
• Arguments 'observation' & 'llm_output' are required if…

AI-assisted analysis of langchain-ai/langchain@e32fa9a52e (2026-08-14). Data as JSON: /api/errors/fa51a58c5b81199c. Report an issue: GitHub.

Appendix: source

Thrown at libs/core/langchain_core/output_parsers/openai_tools.py:69

    if partial:
        try:
            function_args = parse_partial_json(arguments, strict=strict)
        except (JSONDecodeError, TypeError):  # None args raise TypeError
            return None
    # Handle None or empty string arguments for parameter-less tools
    elif not arguments:
        function_args = {}
    else:
        try:
            function_args = json.loads(arguments, strict=strict)
        except JSONDecodeError as e:
            msg = (
                f"Function {raw_tool_call['function']['name']} arguments:\n\n"
                f"{arguments}\n\nare not valid JSON. "
                f"Received JSONDecodeError {e}"
            )
            raise OutputParserException(msg) from e
    parsed = {
        "name": raw_tool_call["function"]["name"] or "",
        "args": function_args or {},
    }
    if return_id:
        parsed["id"] = raw_tool_call.get("id")
        parsed = create_tool_call(**parsed)  # type: ignore[assignment,arg-type]
    return parsed


def make_invalid_tool_call(
    raw_tool_call: dict[str, Any],
    error_msg: str | None,
) -> InvalidToolCall:
    """Create an `InvalidToolCall` from a raw tool call.

    Args:
        raw_tool_call: The raw tool call.

View on GitHub (pinned to e32fa9a52e)