It logs the text and acknowledges — there is no computation behind it. The value is that calling it forces the reasoning to be produced explicitly instead of skipped, which is a real effect and a very cheap one to add. Zero configuration, zero credentials, one tool definition in your context. Whether the model actually reaches for it depends on your prompting, not on the server.
An MCP server that provides a single `reflect` tool. The assistant calls it with its reasoning, the tool logs that thought process to the server console and acknowledges the checkpoint, and the act of having to write the reasoning down is the point.
- A reflection checkpoint the assistant writes into before or after an action — `reflect`
- Structured reasoning used as a planning step: breaking a task into ordered steps before starting
- Task validation after the fact — what was completed, what was learned, what technical debt was created, what comes next
- A debugging scratchpad — symptom, hypotheses, which were confirmed, the solution and the result
- A decision audit trail, since the reasoning is written rather than implicit
- Stdio by default, with an HTTP mode behind environment variables
No account and no key. Install with `npm install -g mcp-reflection-tool` or run it directly with `npx mcp-reflection-tool`; in Claude Code, `mcp add npx mcp-reflection-tool` configures it in one command, then restart the client. Cursor and Cline configurations are documented. For HTTP mode set `HTTP=true`, with `PORT` overriding the default 8080. Node.js 18+ or Bun. Package `mcp-reflection-tool` 0.1.19 with a binary of the same name.
One command — npx mcp-reflection-tool