MCP Chrome extension: frequent transient API Error 500 during tool calls

Description

When using the Claude in Chrome MCP browser extension for browser automation, API Error 500 responses occur frequently and consistently during tool calls. The errors are transient — the tool calls still complete successfully, but each call has a high chance of returning a 500 alongside the successful result.

Environment

• Claude Code on Windows (MSYS2/Git Bash)
• Claude in Chrome browser extension
• MCP tools: mcp__claude-in-chrome__* (navigate, form_input, get_page_text, computer, find, read_page, tabs_context_mcp, etc.)

Reproduction

1. Start a Claude Code session with the Chrome extension connected
2. Perform any sequence of browser automation tool calls (navigate, read_page, form_input, etc.)
3. Observe that a significant portion of calls return API Error 500 alongside successful results

The errors occur across all tool types — not specific to any one MCP tool. They happen intermittently but very frequently (roughly every 2-4 calls in a session).

Error format

API Error: 500 {"type":"error","error":{"type":"api_error","message":"Internal server error"},"request_id":"req_..."}

Sample request IDs

• req_011CYV8YWiHwJchWxE324LWo
• req_011CYV9NQX4oK8QMUM96doRv
• req_011CYV9fp7cR8TDmJZxa4z9c
• req_011CYVAfx3DwUsTmi4zAh7Ev
• req_011CYV9eN4fzTRFXerki4bWJ

Impact

• Slows down multi-step browser automation workflows significantly
• Each error triggers a visible error message in the CLI, causing confusion
• Forces users to repeatedly say "continue" to push past errors
• Makes browser automation sessions take 2-3x longer than necessary

Expected behavior

Tool calls should complete without 500 errors. If the underlying action succeeds (which it does), the response should not include an error.

🤖 Generated with Claude Code