[MCP] Problem with functions_get_logs when using Anthropic models in Antigravity

[paul-van-cotthem]

Summary

In Google Antigravity, there is a critical execution conflict between the Firebase MCP server and Anthropic Claude models (Sonnet/Opus). When Firebase MCP Tool #15 (functions_get_logs) is enabled, Claude models fail to execute any prompt, whereas Google Gemini models continue to function correctly.

Workaround:
Disabling Tool #15 (functions_get_logs) in the Firebase MCP management settings resolves the issue immediately.

Bug Info

• Affected Tool(s): Agent in Antigravity, with Claude Sonnet or Opus models selected.
• MCP Client: Google Antigravity 1.13.3
• Operating System: Windows 11

Steps to Reproduce

Open Google Antigravity.
Open the Agent sidebar (Ctrl+L).
Click the 3-dots icon (top right) > MCP servers.
Click Manage MCP Servers.
Select the Firebase MCP.
Ensure Tool #15: “functions_get_logs” is toggled ON.
Switch the model to Claude Sonnet or Claude Opus.
Send any prompt via the Agent.
Observe the error: “Error: Agent execution terminated due to error”.
Ensure Tool #15: “functions_get_logs” is toggled OFF.
Switch the model to Gemini 3 Flash or Pro.
Send any prompt via the Agent.
Observe answer generated correctly.

[joehan]

This is quite a strange issue - to set expectations, I'm not sure how to tackle it immediately and the problem may well lie outside of this repo. Regardless, more logs is probably the first thing we need to understand why this might happen. If there is a firebase-debug.log from the run, please share it here. Addtionally, would you be able to share the Antigravity diagnostic logs for when the error happens?

Before you share, make sure to review the diagnostic for any info that you don't want to share publicly - I'm not an Antigravity expert, so I don't know off the top of my head what will be in there.

[LumiKitten]

I've been investigating this issue and have some additional context.

From Antigravity's output logs, when functions_get_logs is enabled and Claude is selected, the API requests to https://daily-cloudcode-pa.googleapis.com/v1internal:streamGenerateContent return 400 Bad Request immediately — before any model output is generated. Disabling only this tool resolves the issue.

Possible direction:

After auditing all Firebase MCP tool files, list_projects uses a similar pattern that works:

• list_projects: z.number().min(1).default(20) ✅
• get_logs: z.number().int().min(1).max(1000).default(50) ❌

The additions of .int() and .max() may be triggering the issue. get_logs also uniquely uses z.enum([...]).default("desc") — no other tool puts .default() on an enum.

[paul-van-cotthem]

It seems to have been resolved in the latest update of Antigravity. Thank you for looking into this.

…

________________________________ From: Joe Hanley ***@***.***> Sent: Wednesday, January 7, 2026 7:42:39 PM To: firebase/firebase-tools ***@***.***> Cc: Paul Van Cotthem ***@***.***>; Author ***@***.***> Subject: Re: [firebase/firebase-tools] [MCP] (Issue #9708) [https://avatars.githubusercontent.com/u/4635763?s=20&v=4]joehan left a comment (firebase/firebase-tools#9708)<#9708 (comment)> This is quite a strange issue - to set expectations, I'm not sure how to tackle it immediately and the problem may well lie outside of this repo. Regardless, more logs is probably the first thing we need to understand why this might happen. If there is a firebase-debug.log from the run, please share it here. Addtionally, would you be able to share the Antigravity diagnostic logs for when the error happens? Screenshot.2026-01-07.at.10.37.23.AM.png (view on web)<https://github.com/user-attachments/assets/a84cf5f9-2f69-4006-86cf-b853410b6f16> Before you share, make sure to review the diagnostic for any info that you don't want to share publicly - I'm not an Antigravity expert, so I don't know off the top of my head what will be in there. — Reply to this email directly, view it on GitHub<#9708 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AXJLRSBTMCVPK5GHL2QUXT34FVHR7AVCNFSM6AAAAACQ6O35IGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOMRQGIZTQMRSGM>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[joehan]

Nice, glad that this seems to be fixed on AGYs side. Will close for now, but if this comes up again we'll go try @LumiKitten's suggestions