Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@pr-reviewer-mcp Review PR #12 on myorg/my-repo"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An MCP server exposing AI-powered GitHub PR review as tools, adapted from codewarrior (a webhook-driven GitHub App) into an on-demand tool server.
The original app reviews automatically when a PR opens — event-driven. This version reviews when explicitly asked, by you or by an agent — tool-driven. Same review pipeline (LangChain + Gemini), different trigger model and a different auth model (PAT instead of GitHub App installation tokens, since there's a human in the loop at call time instead of a webhook).
Related MCP server: AI-PR-Review-MCP
Tool | Description |
| List open PRs for a repo |
| Fetch the raw diff, no AI |
| AI review of a single file in a PR |
| Full AI review: all files in parallel + synthesized summary |
| Post a review as a PR comment |
npm install
cp .env.example .envFill in .env:
GITHUB_TOKEN — fine-grained PAT with Pull requests (read/write) and
Contents (read) permissions on the repos you want to review
GOOGLE_API_KEY — from Google AI Studio
npm startThis starts the server on stdio. It won't print anything to stdout (that channel is reserved for the MCP protocol) — status logs go to stderr.
Add to your MCP config (claude_desktop_config.json or equivalent):
{
"mcpServers": {
"pr-reviewer": {
"command": "node",
"args": ["/absolute/path/to/pr-reviewer-mcp/src/index.js"],
"env": {
"GITHUB_TOKEN": "your-token-here",
"GOOGLE_API_KEY": "your-key-here"
}
}
}
}Restart your IDE, then you can ask things like:
"Review PR #12 on srivastava-aryan/dsa-tracker"
and it will call review_pr, decide whether to post it, and call
post_review_comment if you confirm.
Auth: github-auth.js now uses a personal access token via
@octokit/rest directly, instead of @octokit/auth-app + installation
tokens. No webhook server, no Express, no smee-client.
Trigger: no webhook listener. Review happens when a tool is called.
Logging: console.log → console.error in the review pipeline,
since stdout is reserved for MCP protocol messages on stdio transport.
Everything else — prompts, parallel file review, synthesis step — is
unchanged from the original reviewer.js and diff-fetcher.js.
Swap stdio transport for HTTP/SSE if you want to host this remotely instead of running it locally per-user
Add a list_recent_reviews tool backed by a small DB, so repeated
reviews on the same PR don't re-run from scratch
Add resource support (server.resource) to expose PR diffs as
browsable MCP resources rather than only tool outputs
Fetch the raw file-by-file diff for a pull request, with no AI review applied.
Inputs: repo (required), owner (required), pull_number (required).
List open pull requests for a GitHub repo, so you can decide which to review.
Inputs: repo (required), owner (required).
Post a review (typically the output of review_pr) as a comment on the pull request.
Inputs: body (required), repo (required), owner (required), pull_number (required).
Run an AI code review on a single file within a pull request.
Inputs: repo (required), owner (required), filename (required), pull_number (required).
Run a full AI code review on a pull request: fetches the diff, reviews every changed file in parallel, and returns a synthesized markdown review.
Inputs: repo (required), owner (required), pull_number (required).
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 5 tools
Each tool targets a distinct operation: listing PRs, fetching diffs, reviewing individual files, reviewing full PRs, and posting comments. No functional overlap.
All tool names follow a consistent verb_noun pattern in snake_case (list_open_prs, get_pr_diff, review_file, review_pr, post_review_comment), making them predictable.
Five tools is appropriate for a focused PR review server—enough to cover core functionality without being excessive or thin.
Covers the full review workflow (list, diff, review file, review PR, post comment). Lacks tools for approval or dismissal, but those are outside the stated scope.
An MCP server that gives your AI access to the source code and docs of all public github repos
An MCP server that integrates with Discord to provide AI-powered features.
Driflyte MCP server which lets AI assistants query topic-specific knowledge from web and GitHub.
