Skip to main content
Glama
DevDs1989
by DevDs1989
README.md
<div align="center">

<img width="1160" alt="t-rush-mcp" src="./trush-mcp.png" />

# t-rush-mcp

**Speedrun your technical debt, now powered by AI.**

An MCP (Model Context Protocol) server that exposes the power of t-rush directly to AI coding assistants like Antigravity, Claude, and others. Let your AI agents find, prioritize, and fix TODO · FIXME · BUG comments across your codebase.

[![npm version](https://img.shields.io/npm/v/@devds1989/t-rush-mcp?color=black&style=flat-square)](https://www.npmjs.com/package/@devds1989/t-rush-mcp)
[![license](https://img.shields.io/github/license/DevDs1989/trush-mcp?color=black&style=flat-square)](./LICENSE)

</div>

---

## Why t-rush-mcp?

AI coding agents are great at fixing bugs, but they often struggle to know *what* to work on when left unattended. `t-rush-mcp` bridges this gap by exposing your codebase's technical debt directly to the AI through standard MCP tools.

Agents can now query your project for open TODOs, filter them by keyword, prioritize the oldest debt, and even increment your personal t-rush completion streak when they successfully resolve them.

### 💻 Want to fix TODOs manually?
Try the **[t-rush CLI](https://github.com/DevDs1989/trush)** to speedrun your own TODOs! Pick a bug, start a timer, fix the code, and build a massive completion streak yourself.

---

## Features

- **`find_todos`**: Feeds the agent all open TODO/FIXME comments so it can rank them by age and severity natively.
- **`search_todos`**: Allows the agent to fuzzy-search existing comments to find specific tasks.
- **`resolve_todo`**: Verifies the comment has been removed from the file and automatically increments the t-rush streak.
- **`scan_todos_raw`**: Returns an unranked, raw list of all tech debt in a repository.
- **`get_streak_status`**: Exposes the user's current streak and stats to the agent.
- **`aggregate_debt`**: Summarizes the total debt across multiple local repositories.

---

## Install & Configuration

### 1. Via npm (Global Install)

You can install the MCP server globally on your machine using npm:

```bash
npm install -g @devds1989/t-rush-mcp
```

### 2. Manual Configuration

To add this server manually to your client's configuration file (e.g. `claude_desktop_config.json`):

```json
{
  "mcpServers": {
    "t-rush": {
      "command": "npx",
      "args": ["-y", "@devds1989/t-rush-mcp"],
      "eager": true
    }
  }
}
```

Or test it directly via npx:

```bash
npx -y @devds1989/t-rush-mcp
```

### 3. Claude Code

To add this server directly to Anthropic's Claude Code CLI:

```bash
claude mcp add t-rush -- npx -y @devds1989/t-rush-mcp
```

**Requirements:** Node.js 18+

---

## How it works

`t-rush-mcp` uses the shared `@devds1989/trush-core` logic to scan your codebase using the exact same parsers as the `t-rush` CLI. All streak updates and history modifications made by the AI agent are instantly synchronized with your local `~/.t-rush/data.json` database.

When an AI agent resolves a TODO, it increments your streak—teamwork!

---

## Supported languages

t-rush detects `TODO` and `FIXME` in all common comment styles:

| Style | Languages |
|---|---|
| `//` | JavaScript, TypeScript, Go, Rust, C, C++, Java, Kotlin, Swift, Dart |
| `#` | Python, Ruby, Shell, YAML, R, Perl, Elixir, Crystal |
| `--` | SQL, Lua, Haskell, Ada |
| `%` | Erlang, LaTeX |
| `;` | Lisp, Clojure, Assembly |
| `*` | Inside `/* */` block comments |

---

## Contributing

Contributions are welcome. Please open an issue before submitting a large PR.

```bash
git clone https://github.com/DevDs1989/trush-mcp
cd trush-mcp
npm install
npm run build
```

---

## License

[MIT](./LICENSE) © Dev

TDQS

A3.7/5.0

Scored across 6 tools

Disambiguation4/5

Most tools have clearly distinct purposes: find_todos ranks priorities, scan_todos_raw returns an unranked list, search_todos does fuzzy searches, resolve_todo marks completion, and get_streak_status/aggregate_debt handle stats. The main potential confusion is between find_todos and scan_todos_raw, as both scan for TODOs, though the ranking distinction helps.

Naming Consistency4/5

All tool names use lowercase snake_case with leading verbs (find, resolve, scan, search, get, aggregate), forming a consistent pattern. Minor deviations exist: resolve_todo is singular while most others are plural, and scan_todos_raw appends 'raw' as a modifier, slightly breaking the simple verb_noun structure.

Tool Count5/5

Six tools are well-scoped for a tech debt tracking server. Each tool serves a distinct aspect—scanning, ranking, searching, resolving, streak stats, and cross-repo aggregation—so the count feels appropriate without being sparse or bloated.

Completeness4/5

The tool surface covers the core lifecycle: scanning and ranking TODOs, searching for specific items, resolving them, and tracking streaks across repositories. Minor gaps exist, such as no ability to reopen a resolved TODO or add a TODO, but these are acceptable limitations for a tech debt scanner.

Maintenance

ActivityStale
ResponsivenessNo issues