Schema validation error with Google/Gemini models #8036

Description

When using OpenCode with Google Gemini models, I'm encountering a 400 validation error related to the memory_rate_memory tool's schema.

Error Message

Invalid value at 'request.tools[0].function_declarations[33].parameters.properties[1].value.enum[0]' (TYPE_STRING), -1
Invalid value at 'request.tools[0].function_declarations[33].parameters.properties[1].value.enum[1]' (TYPE_STRING), 0
Invalid value at 'request.tools[0].function_declarations[33].parameters.properties[1].value.enum[2]' (TYPE_STRING), 1

Details

The memory_rate_memory tool (from the mcp-memory-service MCP server) defines its rating parameter with numeric enum values:

"rating": {
  "description": "Quality rating: -1 (thumbs down), 0 (neutral), 1 (thumbs up)",
  "type": "number",
  "enum": [-1, 0, 1]
}

When OpenCode passes this tool definition to the Google/Gemini API, the API rejects it with a 400 error, indicating it expects string enum values instead of numeric ones.

Context

• mcp-memory-service is a popular, actively maintained MCP server (1.1k+ stars, Apache 2.0 licensed)
• It's compatible with 13+ AI applications including Claude Desktop, Claude Code, VS Code, Cursor, etc.
• The tool appears to work correctly with other model providers (OpenAI, Anthropic, etc.)
• This issue appears specific to Google/Gemini's stricter API validation requirements

Environment

• OpenCode version: latest
• Model: gemini-3-pro-low
• Platform: Linux

Additional Context

The error occurs when trying to use tools from mcp-memory-service with Google Gemini models in OpenCode. Other model providers seem to accept the same tool schema without issues.