# codereviewbuddy

**Category:** 🔄 Version Control  
**Repository:** https://github.com/detailobsessed/codereviewbuddy  
**GitHub Stars:** 0  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/codereviewbuddy

## Description
MCP server that helps AI coding agents interact with AI code reviewers on GitHub PRs

## Tools
Capabilities this server exposes over MCP:

- **summarize_review_status** — Lightweight stack-wide overview — start here
- **triage_review_comments** — Only actionable inline threads needing attention
- **get_thread** — Full thread details by node ID — use after triage for conversation history
- **reply_to_comment** — Reply to inline threads (`PRRT_`), PR-level reviews (`PRR_`), or bot comments (`IC_`)
- **diagnose_ci** — Diagnose CI failures — finds the failed run, jobs, steps, and error lines in one call
- **check_ci_status** — Lightweight CI pass/fail/pending check for a PR — use before merging
- **stack_activity** — Chronological activity feed across a PR stack with a `settled` flag
- **list_recent_unresolved** — Scan recently merged PRs for unresolved review threads
- **review_pr_descriptions** — Analyze PR descriptions for quality issues (empty body, boilerplate, missing linked issues)
- **show_config** — Show active configuration with human-readable explanation

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

```json
"mcpServers": {
  "codereviewbuddy": {
    "command": "uvx",
    "args": ["codereviewbuddy"]
  }
}
```

## Documentation

## What codereviewbuddy MCP server does

The codereviewbuddy MCP server gives AI coding agents structured access to GitHub pull-request review work. It covers review status, actionable inline comments, thread history, replies, CI results, stacked-PR activity, and selected pull-request hygiene checks.

A typical workflow starts with `summarize_review_status` for a lightweight view across a PR stack. The agent can then use `triage_review_comments` to find inline threads that still need action, followed by `get_thread` for complete conversation history. `reply_to_comment` supports inline review threads, PR-level reviews, and bot issue comments by routing each identifier type to the corresponding GitHub API operation.

The server also handles review follow-up outside the current PR. `list_recent_unresolved` scans recently merged pull requests for review threads that were not resolved, while `review_pr_descriptions` checks for issues such as an empty body, boilerplate content, or missing linked issues.

## How it works

The codereviewbuddy MCP server runs through FastMCP and returns typed Pydantic models with JSON Schema rather than unstructured tool output. It uses the local `gh` CLI for GitHub authentication, so a personal access token or `.env` file is not required by the server. Before starting it, install and authenticate GitHub CLI with `gh auth login`.

Project context is detected from MCP roots sent by the client. This allows separate client windows to work with different projects without adding a repository environment variable. Tool responses can include `next_steps`, explanations for empty results, direct GitHub comment URLs, and recovery guidance for authentication, rate-limit, missing-resource, workspace, GraphQL, or configuration errors.

The server also exposes a read-only review resource at `pr://{owner}/{repo}/{pr_number}/reviews`. Its prompts include `review_stack`, `pr_review_checklist`, and `ship_stack` for guided review and pre-merge workflows.

## Setup and configuration

Python 3.14 or newer, `uv`, and an authenticated GitHub CLI installation are required. The simplest launch command is:

```bash
uvx codereviewbuddy@latest
```

Permanent installation is available with `uv tool install codereviewbuddy`. The project includes an installer for Claude Desktop, Claude Code, Cursor, and Windsurf, plus a command that prints or copies generic MCP JSON configuration. Restart the client after changing its MCP configuration.

Configuration is optional. Settings use `CRB_*` environment variables in the MCP client's `env` block, with double underscores separating nested settings. Supported examples include `CRB_OWNER_LOGINS`, `CRB_PR_DESCRIPTIONS__ENABLED`, and `CRB_SELF_IMPROVEMENT__ENABLED`.

## Tools and capabilities

The codereviewbuddy MCP server exposes these tools:

- `summarize_review_status` and `stack_activity` provide overview and chronological stack information; `stack_activity` includes a `settled` flag.
- `triage_review_comments` filters for actionable inline threads and supplies GitHub URLs.
- `get_thread` retrieves a full thread by node ID.
- `reply_to_comment` replies to inline, PR-level, or bot comments.
- `diagnose_ci` identifies failed runs, jobs, steps, and error lines in one call.
- `check_ci_status` reports whether a PR's CI is passing, failing, or pending.
- `list_recent_unresolved` finds unresolved review threads on recently merged PRs.
- `review_pr_descriptions` checks pull-request description quality.
- `show_config` displays active settings with explanations.

FastMCP also provides progress reporting where supported by the client, plus middleware for error handling, timing logs, and request/response logging.

## Limitations and notes

The server depends on GitHub CLI authentication and a compatible local Python runtime. It is designed around GitHub's pull-request review infrastructure, so its comment operations use GitHub identifier formats such as `PRRT_`, `PRR_`, and `IC_`. Client support for progress display, structured output, tool filtering, or other FastMCP features may vary.

When using a locally checked-out source version, launch it with `uv run --directory` so the client uses the repository's dependencies. If FastMCP reports a missing `fastmcp.server.tasks.routing` module, the documented remedies are to use `codereviewbuddy@latest`, run from the source checkout with `uv`, or reinstall the tool to refresh cached dependencies.

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