API Error: Unable to connect to API (ConnectionRefused) Using Claude Pro Max Subscription

[Infizzle]

Unable to start projects, receiving various forms of rate limit / api errors. I do not think it is my authentication as I have other working projects locally using CC CLI. Running the latest version of autoforge as of today.

[GRajuPy]

Yes, consistently hitting rate limit - even though other CC CLI projects work w/o any issues, and current usage is near 0% for the current session and weekly limits. I am using a monthly subscription key.

[Javier-Viramontes]

I am encountering the same error:
Error during agent session: Unknown message type: rate_limit_event.

However, in my case, the project stopped completely.

[Bortus-AI]

ditto

[kenmoini]

Also seeing this issue, CC CLI, VSCode, etc work fine - even upgraded my account with more usage. Autoforge will not get out of this rate_limit_event.

[Infizzle]

FWIW v0.1.3 on another machine seems to be working, using the same CC CLI auth. I just upgraded it to v0.1.13 and it seems to be working.

[Infizzle]

tried using "autoforge --repair" on the machine that wasn't working, but that did not help.

[GRajuPy]

I asked Claude to find root clause. It provided following information:
Here's what was happening:

Root cause: The Claude CLI added a new rate_limit_event message type, but the installed claude_agent_sdk version doesn't handle it — falling through to case _ which raises MessageParseError: Unknown message type: rate_limit_event. This exception crashed the receive_response() loop mid-session.

Two-part fix:

1. message_parser.py (both venvs) — Add a case "rate_limit_event" handler that returns a SystemMessage instead of raising an error. This stops the crash and lets the session continue uninterrupted (the CLI handles the actual rate-limit waiting internally).
2. agent.py — Add a SystemMessage handler in the receive loop that prints a user-friendly message when a rate_limit_event is received (e.g., [Rate limit event - CLI will retry in 60s...]), so you know what's happening without alarming error output.

This fixed the error for me.

[Javier-Viramontes]

I asked Claude to find root clause. It provided following information: Here's what was happening:

Root cause: The Claude CLI added a new rate_limit_event message type, but the installed claude_agent_sdk version doesn't handle it — falling through to case _ which raises MessageParseError: Unknown message type: rate_limit_event. This exception crashed the receive_response() loop mid-session.

Two-part fix:

1. message_parser.py (both venvs) — Add a case "rate_limit_event" handler that returns a SystemMessage instead of raising an error. This stops the crash and lets the session continue uninterrupted (the CLI handles the actual rate-limit waiting internally).
2. agent.py — Add a SystemMessage handler in the receive loop that prints a user-friendly message when a rate_limit_event is received (e.g., [Rate limit event - CLI will retry in 60s...]), so you know what's happening without alarming error output.

This fixed the error for me.

Hello, it is working for me as well. Thank you!

[granowskiii]

I asked Claude to find root clause. It provided following information: Here's what was happening:

Root cause: The Claude CLI added a new rate_limit_event message type, but the installed claude_agent_sdk version doesn't handle it — falling through to case _ which raises MessageParseError: Unknown message type: rate_limit_event. This exception crashed the receive_response() loop mid-session.

Two-part fix:

1. message_parser.py (both venvs) — Add a case "rate_limit_event" handler that returns a SystemMessage instead of raising an error. This stops the crash and lets the session continue uninterrupted (the CLI handles the actual rate-limit waiting internally).
2. agent.py — Add a SystemMessage handler in the receive loop that prints a user-friendly message when a rate_limit_event is received (e.g., [Rate limit event - CLI will retry in 60s...]), so you know what's happening without alarming error output.

This fixed the error for me.

Works for me as well, big thanks!

[Infizzle]

@GRajuPy thanks for helping out! I tried your fix but still no luck. Would you mind sharing any additional details that would help me and perhaps others? Thank you again for your help on this!!

[GRajuPy]

I am using code from the master branch - version 1.3

In the root directory
File: agent.py
Diff details:

97a98,108

        # Handle SystemMessage (rate_limit_event and other system events)
        elif msg_type == "SystemMessage":
            subtype = getattr(msg, "subtype", "")
            if subtype == "rate_limit_event":
                msg_data = getattr(msg, "data", {})
                retry_ms = msg_data.get("retryAfterMs") or msg_data.get("retry_after_ms")
                if retry_ms:
                    print(f"\n[Rate limit event - CLI will retry in {retry_ms / 1000:.0f}s...]", flush=True)
                else:
                    print("\n[Rate limit event - waiting for API to recover...]", flush=True)

---

Another file is message_parser.py; depending on how you installed AutoForge, its location may differ. It could be in the Python site-packages directory, or in the venv/lib/python3.12/site-packages/claude_agent_sdk/_internal/message_parser.py directory, under the autocoder-master folder. (If you have multiple message_parser.py files and are not sure which is getting used, you may want to make the following (or similar) change in all message_parser.py files.)

File: message_parser.py
Diff details:

178a179,184

    case "rate_limit_event":
        return SystemMessage(
            subtype="rate_limit_event",
            data=data,
        )

---

Hope this helps,