# mirror-mcp [Health: Active]

**Category:** 🗣️ Conversational AI  
**Repository:** https://github.com/toby/mirror-mcp  
**GitHub Stars:** 15  
**npm Downloads (last month):** 187  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/mirror-mcp

## Description
Mirror gives LLMs the ability to self-reflect with MCP sampling.

## Tools
Capabilities this server exposes over MCP:

- **reflect** — Enable targeted LLM self-reflection with customizable prompts for specialized analysis. Use custom system_prompt and user_prompt parameters to get focused, expert-level reflections instead of generic responses. Particularly effective for domain-specific analysis, critical thinking, coaching perspectives, and structured output formats.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "mirror-mcp": {
    "command": "npx",
    "args": ["-y","mirror-mcp"]
  }
}
```

## Documentation

## What mirror-mcp MCP server does

The mirror-mcp MCP server adds one MCP tool, `reflect`, for model-directed self-questioning. An LLM supplies a question and can include prior context, then receives a response generated through MCP sampling. The result contains the reflection and metadata such as token usage and elapsed reflection time.

The tool is intended for reasoning checks rather than access to an external application or data source. Typical prompts can ask the model to assess confidence, identify weaknesses, inspect assumptions, or consider other viewpoints.

## How it works

The calling MCP client sends the reflection request to mirror-mcp, which passes the self-directed question to the MCP sampling interface. Sampling options control response generation, while the returned result is formatted for the client to consume.

Without custom instructions, the tool uses built-in guidance oriented around strengths, weaknesses, assumptions, and alternative perspectives. A caller can instead provide a system prompt to establish a role or perspective and a user prompt to specify the desired structure or focus. This makes mirror-mcp suitable for tasks such as critical review, coaching-style analysis, and iterative problem-solving.

The `max_tokens` option defaults to 500, and `temperature` defaults to 0.8. The tool also accepts a separate `context` value so the reflection can refer to information from the current reasoning process.

## Setup and configuration

The mirror-mcp MCP server can run without a local installation by using:

```bash
npx mirror-mcp
```

For a persistent npm installation, install the package globally with `npm install -g mirror-mcp`. MCP hosts using stdio can launch it with the following configuration:

```json
{
  "type": "stdio",
  "command": "npx",
  "args": ["mirror-mcp@latest"]
}
```

VS Code has a documented one-click installation path. Developers running from source need Node.js 18 or newer, then clone the repository, install dependencies, build the project, and start it. Development also provides a `npm run dev` command and a test command.

## Tools and capabilities

`reflect` accepts these arguments:

- `question` — required string containing the self-directed question.
- `context` — optional supporting information.
- `system_prompt` — optional instruction defining the reflection role or perspective.
- `user_prompt` — optional instruction controlling response format, structure, or focus.
- `max_tokens` — optional response limit, defaulting to 500.
- `temperature` — optional sampling setting, defaulting to 0.8.

The response includes a `reflection` string and metadata with token usage and reflection time.

## Limitations and notes

Reflection quality depends on the model and MCP client providing sampling support. The server does not independently verify the model’s conclusions; its output is another model-generated perspective. Custom prompts replace or shape the default reflection guidance, so callers should specify the desired review method when consistent structure matters.

The documented deployment uses stdio. No environment variables or authentication settings are specified. The project is distributed under the MIT license.

_Full upstream README: https://allmcps.com/mcp/mirror-mcp/readme_