# Dear User [Health: Active]

**Category:** 🗣️ Conversational AI  
**Repository:** https://github.com/bleedmode/dearuser  
**GitHub Stars:** 0  
**npm Downloads (last month):** 270  
**Views:** 1  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/dear-user

## Description
Tells you how you and your Claude agent actually work together. Local-only, no API keys.

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

```json
"mcpServers": {
  "dear-user": {
    "command": "npx",
    "args": ["-y","@poisedhq/dearuser-mcp@latest"]
  }
}
```

## Documentation

## What Dear User MCP server does

The Dear User MCP server evaluates how an AI coding setup is configured and how effectively it supports collaboration. It works with Claude Code and Claude Desktop, and its AGENTS.md support also covers setups used with Cursor, Codex, Aider, Cline, and Zed. There is no graphical interface, account requirement, or mandatory cloud service.

The main reports are collaboration, security, health, and Wrapped. Collaboration analysis assigns a persona, scores seven categories from 0 to 100, identifies friction patterns, and produces prioritized recommendations. Security analysis looks for exposed credentials, prompt-injection surfaces, and contradictory rules. Health analysis checks structural issues such as orphaned scheduled work, overlapping skills, stale or dead hooks, and registration gaps. Wrapped turns selected scores and counts into an optional shareable card.

## How it works

Dear User MCP server reads local configuration and project-support files, including CLAUDE.md or AGENTS.md, memory files, skills, hooks, and session metadata. It reads session counts and lengths rather than conversation text. Scanning, parsing, scoring, secret detection, conflict detection, persona detection, and recommendation generation happen locally.

Results are saved in `~/.dearuser/dearuser.db` as a SQLite database. Later history queries can show previous reports and score changes without running another scan. A feedback loop records recommendation status so later reports can indicate which suggestions were implemented and how the score changed.

The default operation is local-only. The `share_report` action is a separate opt-in path for Wrapped results: it anonymizes paths, removes email addresses, and redacts detected secrets before uploading a card to `dearuser.ai`. The `feedback` tool sends only the message supplied to it to the project’s Supabase inbox.

## Setup and configuration

Install the MCP package with the Claude Code command:

```bash
claude mcp add --scope user dearuser -- npx -y @poisedhq/dearuser-mcp@latest
```

For Claude Desktop, add an MCP entry that runs `npx` with `@poisedhq/dearuser-mcp` as its argument. The repository also provides installation guidance for Cursor, Windsurf, Cline, and Zed. Restart the client after installation so the tools become available.

Optional slash commands can be installed with `dearuser-install-skills`. Sharing Wrapped cards requires `DEARUSER_SUPABASE_URL` and `DEARUSER_SUPABASE_SERVICE_KEY`; without them, sharing fails while the other functions continue to work. No API key is needed for local reports.

## Tools and capabilities

The Dear User MCP server exposes these tools and helpers:

- `collab`: generates a collaboration report, persona, score, friction findings, and recommendations.
- `security`: scans supported agent files and related resources for secrets, injection surfaces, and rule conflicts.
- `health`: checks setup structure, scheduled tasks, skills, hooks, and related consistency issues.
- `wrapped`: creates local collaboration statistics for a Wrapped-style card.
- `share_report`: optionally uploads an anonymized Wrapped card and can set an expiration time.
- `history`: retrieves earlier reports and trends without rescanning.
- `onboard`: guides a new user through seven setup steps.
- `implement_recommendation` and `dismiss_recommendation`: manage pending suggestions.
- `feedback` and `help`: send a note or list available functions.

Conflict detection can identify semantic contradictions, not only rules with matching words. The secret scanner covers provider-specific keys, private keys, generic environment secrets, and bearer tokens across the supported local files.

## Limitations and notes

The Dear User MCP server does not automatically publish collaboration, security, or health findings. Those reports remain local because their details may expose project or business context. Wrapped sharing is the only report-upload path, and feedback is a separate explicit transmission.

The tool reads files and metadata from the local setup, so its findings depend on what is present and accessible. Sharing requires the two Supabase environment variables, while local scanning does not. The documentation names Claude Code, Claude Desktop, Cursor, Windsurf, Cline, and Zed installation paths; Aider is mentioned as an AGENTS.md-compatible environment rather than an installation target in the provided instructions.

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