# devcoach [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/UltimaPhoenix/dev-coach  
**GitHub Stars:** 5  
**npm Downloads (last month):** 4647  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/devcoach

## Description
Progressive in-context technical coaching in your AI agent — one local, free lesson per task.

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

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

## Documentation

## What devcoach MCP server does

The devcoach MCP server adds short technical lessons to an AI-assisted development workflow. After a task is completed, it checks whether coaching is due, determines whether the work is technical, selects a topic and depth, and records the lesson. The intended result is a brief prompt tied to the work that just happened rather than a separate, generic tutorial.

All coaching data remains on the local machine. The server uses one SQLite database at `~/.devcoach/coaching.db`; the project states that it does not use telemetry, accounts, or outbound calls. This makes it suitable for developers who want local learning history alongside their agent sessions.

## How it works

The devcoach MCP server uses an MCP skill and, where configured, a stop hook. Completed tasks first pass through pacing and rate-limit checks. If a lesson is not due, the workflow stays silent and increments its counter. When coaching is due, the skill checks whether the task involved technical work. Non-technical work is skipped; eligible work results in topic selection, depth calibration, lesson logging, and a lesson card.

The lesson can prompt the developer for feedback such as whether it was understood. The documented flow resets the counter after a lesson and re-arms pacing after a skipped task. This approach is intended to limit interruptions while still making coaching part of normal agent use.

## Setup and configuration

The standard package runs locally as a stdio MCP server and requires Node.js 24 or newer because it uses the embedded `node:sqlite` module. With a compatible client, the documented command is `npx -y devcoach install`; restart the agent afterward. Homebrew installation is available for macOS and Linux and also provides the `devcoach` command-line interface.

Claude Code users can install the provided plugin, which bundles the MCP server, automatic-coaching stop hooks, and skill. The plugin should not be combined with `devcoach install`, because both would register the hooks. Claude Desktop users can instead install the supplied `.mcpb` extension; that bundle uses Claude Desktop's runtime and does not require Node or a terminal.

The project also lists the server in the official MCP Registry under `io.github.UltimaPhoenix/devcoach`.

## Tools and capabilities

The documented MCP tools and related commands include:

- `log_lesson` records a selected lesson.
- `skip_lesson` handles work that should not receive a lesson.
- `open_ui` opens the web dashboard.
- The CLI supports dashboard access, lesson and statistics views, sharing and importing lessons, and database backup.

Claude Code users can also access the dashboard with `/devcoach:ui`, share or import lessons, and turn a lesson they could not follow into a step-by-step course through `/devcoach:course`. These plugin commands are separate from the core lesson-selection flow.

## Limitations and notes

The server is local and stores its state in a SQLite file on the machine where the agent runs. It is therefore not presented as a hosted, shared coaching service. Automatic lesson delivery depends on the client integration and its hooks or plugin support; simply connecting an MCP client does not imply that every client provides the same automation.

Gemini CLI and Codex CLI are identified as beta integrations in the README, and the Claude Desktop extension is also marked beta. The Claude Desktop extension is self-signed and installs as an unverified publisher. The project is licensed under AGPL-3.0.

_Full upstream README: https://allmcps.com/mcp/devcoach/readme_