## TL;DR
The server logs thoughts to stderr by default, which lands in your CLI logs. Set `DISABLE_THOUGHT_LOGGING=true` in the server's `env` block. The model still receives thoughts through the tool protocol; only the stderr trace goes quiet.

## The error
```text
(stderr full of sequential thought traces on every tool call)
```

## Fix it
1. **Add the env var to your MCP config:**
   ```json
   {
     "mcpServers": {
       "sequential-thinking": {
         "command": "npx",
         "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"],
         "env": { "DISABLE_THOUGHT_LOGGING": "true" }
       }
     }
   }
   ```
   - Expected: valid JSON.
2. **Restart the client.**
   - Expected: logs are clean; thinking still works in-chat.

## When to use
- You run a CLI client and stderr thought traces pollute logs or CI output.
- You want the reasoning in-context but not in logs.

## When NOT to use
- The model is not thinking at all - that is a client/tool-availability issue.
- You rely on the stderr trace for debugging - keep it on while debugging.

## Compatibility
- @modelcontextprotocol/server-sequential-thinking.

## Root cause
The server is passive: the model does the reasoning and the server just records it. The stderr trace is a debugging convenience that defaults on; the actual thought content flows through the tool result, which is unaffected by the flag.

## Edge cases
- The value must be the string "true" in JSON env.
- If you later need to debug thought flow, remove the flag temporarily.