Skip to main content
Glama
cqfn
by cqfn
README.md
# MCP Server for Aibolit, Java Code Analyzer

[![DevOps By Rultor.com](https://www.rultor.com/b/cqfn/aibolit-mcp-server)](https://www.rultor.com/p/cqfn/aibolit-mcp-server)

[![make](https://github.com/cqfn/aibolit-mcp-server/actions/workflows/make.yml/badge.svg)](https://github.com/cqfn/aibolit-mcp-server/actions/workflows/make.yml)
[![codecov](https://codecov.io/gh/cqfn/aibolit-mcp-server/graph/badge.svg?token=3dr2Il91wJ)](https://codecov.io/gh/cqfn/aibolit-mcp-server)
[![Hits-of-Code](https://hitsofcode.com/github/cqfn/aibolit-mcp-server)](https://hitsofcode.com/view/github/cqfn/aibolit-mcp-server)
[![License](https://img.shields.io/badge/license-MIT-green.svg)](https://github.com/cqfn/aibolit-mcp-server/blob/master/LICENSE.txt)

If you use AI agents, such as [Claude Code], [Cursor], or [Windsurf],
  for code refactoring, you may enjoy using this [MCP] server.
Your AI agent, when you ask it to "make code better,"
  may wonder what exactly needs to be improved.
Sadly, it may often overlook important problems.
AI agents, by design, pay more attention to cosmetic issues,
  which are "low-hanging fruits" for them.
This [MCP] server will give your agent a hint:
  what is the **most critical design issue** in the code.
Then, the agent will refactor it and fix the issue.

First, install [Node], [Npm], [Python], [Pip], and [aibolit]:

```bash
aibolit --version
```

Then, add this [MCP] server to [Claude Code]
(or simply edit `~/claude.json`, but it's not recommended):

```bash
claude mcp add aibolit npx aibolit-mcp-server@0.0.6
```

Then, restart [Claude Code] and ask it something along these lines:
"Find the most critical design issue in my code base and fix it."

## How to Contribute

To test this project, simply run the following commands
(you'll need [Node] 18+, [Npm], and [GNU make] installed):

```bash
npm install
make
```

If everything builds correctly after your changes, submit a pull request.

[MCP]: https://modelcontextprotocol.io/
[Npm]: https://www.npmjs.com/
[Node]: https://nodejs.org/en
[aibolit]: https://github.com/cqfn/aibolit
[Claude Code]: https://github.com/anthropics/claude-code
[Cursor]: https://www.cursor.com/
[Windsurf]: https://windsurf.com/editor
[GNU make]: https://www.gnu.org/software/make/
[Python]: https://www.python.org/
[Pip]: https://pypi.org/project/pip/

TDQS

A3.9/5.0

Scored across 1 tool

Disambiguation5/5

With only one tool, there is no possibility of ambiguity or overlap. The tool's purpose is clearly defined.

Naming Consistency5/5

Single tool, so no inconsistency. The name follows a verb_noun pattern (find_issue).

Tool Count3/5

One tool for a specific purpose (finding design issues in one file) is borderline but acceptable for a narrow scope.

Completeness2/5

The server only covers a single operation (analyze one file for one issue). Lacks features like batch processing, multiple issue types, or suggestions, leaving significant gaps for a design review tool.

Maintenance

ActivityInactive
ResponsivenessUnresponsive