@nochinxx
基本情報
Swipe-to-review GitHub PRs with AI context. Paste a repo URL, get open PRs as a card stack with risk scores, AI summaries, similar past changes, and contributor history. Works as a Claude MCP plugin — review PRs directly from Claude Desktop or Cursor without a browser. 12 tools i
以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。
{
"mcpServers": {
"swipr": {
"command": "npx",
"args": [
"-y",
"supergateway",
"--streamableHttp",
"https://v0-swipr-build.vercel.app/api/mcp"
]
}
}
}Resolve `owner/repo#number` → internal ID
Full PR data: files, patches, risk score, contributor stats, cached AI summary
0–100 heuristic score with reasons
Past PRs in the same repo with semantic similarity
PR count, merge rate, first contribution date
Raw file content at HEAD
Search patches for usages of a function name
Test files that likely cover the changed code
Recent contributors to a file path
File content at an arbitrary git ref
Capture approve / changes / skip
End-of-session stats
Swipr is a swipe-to-review interface for GitHub pull requests, providing AI-generated context such as risk scores, contributor history, similar past changes, and test coverage. It works as both a web app and an MCP (Model Context Protocol) server, enabling human reviewers and AI agents (e.g., Claude) to triage PRs quickly.
Add the MCP server to Claude Desktop by editing claude_desktop_config.json with the provided npx command using Supergateway and the hosted endpoint. Once configured, ask Claude to analyze PRs via tools like lookup_pr, analyze_pr, or find_similar_changes. The web app at the demo URL lets you paste a public repo, then swipe right (approve), left (request changes), or down (skip) with keyboard shortcuts (J, F, Space).
analyze_pr, risk_score, find_related_tests)lib/scoring.ts)Both surface the same PR context (risk, similar changes, contributor history, tests). The web app provides a visual swipe interface for humans; the MCP server lets AI agents like Claude access the same data programmatically.
Self-hosted Swipr stores PR data in Neon Postgres with pgvector. The free-tier Neon instance (512 MB) holds roughly 5,000 PRs (~100 KB each). Ingestion is capped at 100 open PRs per repo.
The risk heuristic is defined in `lib/scoring.ts` as plain TypeScript (no ML). You can add or adjust rules, e.g., increase score when auth files are touched. Color thresholds (green/yellow/red) at 40 and 70 are set in `app/swipe/_components/view-helpers.ts`.
The hosted MCP server uses no authentication — the URL is public. All 12 tools are pure database reads; no AI API calls are made, so no AI credits are consumed. The chat
The system of record for AI-written software. A persistent graph of entities, relationships, changes, and provenance, so humans and AI agents see what a change touches before it merges. Beside Git today.
A Git MCP server for AI agents. STDIO & Streamable HTTP.
a mcp server to manage github project's functionality
Node.js/TypeScript MCP server for Atlassian Bitbucket. Enables AI systems (LLMs) to interact with workspaces, repositories, and pull requests via tools (list, get, comment, search). Connects AI directly to version control workflows through the standard MCP interface.
Interact seamlessly with GitLab repositories to manage merge requests and issues. Fetch details, add comments, and streamline your code review process with ease.
コメント