pr-narrator-mcp
Generates commit messages, PR titles and descriptions, and release notes from git changes, with automatic domain detection for appropriate PR templates.
README
pr-narrator-mcp
<p align="center"> <img src="https://raw.githubusercontent.com/mhaviv/pr-narrator-mcp/main/assets/social-preview.png" alt="PR Narrator — AI-powered commit messages & PR descriptions" width="100%" /> </p>
Generate consistent commit messages, PR content, and release notes automatically.
An MCP server that generates commit messages, PR content (titles, descriptions, and templates), and release notes from your git changes. If your repo doesn't already have a PR template, it auto-detects the domain (mobile, frontend, backend, devops, security, ML) and applies the right one — no config needed.
Install
npx pr-narrator-mcp
Quick Start
Cursor
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"pr-narrator": {
"command": "npx",
"args": ["-y", "pr-narrator-mcp"],
"env": {
"BASE_BRANCH": "develop",
"TICKET_PATTERN": "[A-Z]+-\\d+"
}
}
}
}
Claude Desktop
Add to your Claude Desktop config:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json - Linux:
~/.config/Claude/claude_desktop_config.json
{
"mcpServers": {
"pr-narrator": {
"command": "npx",
"args": ["-y", "pr-narrator-mcp"],
"env": {
"BASE_BRANCH": "develop",
"TICKET_PATTERN": "[A-Z]+-\\d+"
}
}
}
}
Windsurf
Add to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"pr-narrator": {
"command": "npx",
"args": ["-y", "pr-narrator-mcp"],
"env": {
"BASE_BRANCH": "develop",
"TICKET_PATTERN": "[A-Z]+-\\d+"
}
}
}
}
VS Code (GitHub Copilot)
Add to .vscode/mcp.json in your project:
{
"servers": {
"pr-narrator": {
"command": "npx",
"args": ["-y", "pr-narrator-mcp"],
"env": {
"BASE_BRANCH": "develop",
"TICKET_PATTERN": "[A-Z]+-\\d+"
}
}
}
}
Other MCP Clients
Any MCP client that supports stdio transport can use this server. The command is:
npx -y pr-narrator-mcp
That's it! No config files needed. All env vars are optional.
Settings
All settings are optional env vars in MCP JSON:
| Env Var | What it does | Example |
|---|---|---|
BASE_BRANCH |
Base branch for PRs | develop |
TICKET_PATTERN |
Ticket regex | [A-Z]+-\\d+ |
TICKET_LINK |
Ticket URL template | https://jira.example.com/browse/{ticket} |
PREFIX_STYLE |
Prefix format | capitalized or bracketed |
DEFAULT_REPO_PATH |
Fallback repo path (single-repo workflows) | /Users/me/my-project |
PR_TEMPLATE_PRESET |
Force a PR template preset | mobile, backend, devops |
PR_DETECT_REPO_TEMPLATE |
Enable/disable repo template detection | true (default) or false |
If BASE_BRANCH is not set, it auto-detects from the repo (main, master, develop).
Note on repoPath: All tools accept a repoPath parameter. The AI calling the tool should pass the user's current workspace directory. DEFAULT_REPO_PATH is only a fallback for single-repo workflows.
Tools
PR Generation
generate_pr
Generate a complete PR with title, description, and context for AI enhancement. Automatically resolves the best template for the repo (see PR Templates).
Returns:
title— PR title (placeholder derived from branch/commits — AI should rewrite)description— PR description with sections from the resolved templatepurposeContext— ALL commit titles, ALL commit bullets, test info, file countpurposeGuidelines— Instructions for AI to rewrite title and Purpose from ALL data
Important: Both the title and Purpose are placeholders. The AI must read ALL purposeContext.commitTitles and purposeContext.commitBullets to synthesize a title and description that reflects the full scope of changes.
Optional templatePreset parameter to force a specific template (e.g., mobile, backend).
generate_pr_title
Generate a PR title based on branch info and commits.
generate_pr_description
Generate a PR description with auto-populated sections. Same template resolution as generate_pr but returns only the description. Accepts optional templatePreset and summary parameters.
get_pr_template
Preview the resolved PR template for a repo before generating. Shows which sections will appear based on the repo's template file, domain auto-detection, or configured preset. Useful for understanding what a PR will look like before calling generate_pr.
Returns the template source (repo, preset, auto-detected, or default), detected domain, and each section's visibility based on current branch changes.
Commit Messages
generate_commit_message
Prepare commit message context from staged or unstaged changes.
If nothing is staged, the tool automatically falls back to unstaged working tree changes and provides staging instructions — no need to run git add first just to analyze your changes. The response includes a source field ("staged" or "unstaged") and a hint with the exact git add command to run.
When includeBody is true, the actual diff is provided so the AI can write a meaningful body that describes what changed functionally — not just file type counts.
Two modes:
- With
summaryparam (recommended): Returns a ready-to-use commit message with proper prefix/formatting. AddincludeBody: trueto get diff-based body generation. - Without
summary: Returns placeholder title + diff + guidelines for AI to compose the message and body
validate_commit_message
Check a commit message against configured rules (length, format, capitalization, imperative mood).
Release Notes
generate_changelog
Generate a changelog / release notes from git commit history between two refs (tags, SHAs, or branches).
- Auto-resolves refs — defaults
fromto the latest tag (or initial commit if no tags) andtoto HEAD - Parses commits — detects conventional commit types/scopes, infers types from keywords for non-conventional messages, extracts co-authors and ticket references
- Deduplicates squash-merge artifacts
- Three output formats:
keepachangelog(default),github-release,plain - Three grouping modes:
type(default),scope,ticket - Includes stats — commit count, contributor count, ticket count, and a one-line summary
Returns changelog (formatted markdown), entries (structured data), summary, stats, range, and warnings.
Repository Analysis
analyze_git_changes
Analyze the current repository state: staged changes, branch info, working tree status, and file categorization.
extract_tickets
Find ticket numbers in the branch name and commit messages using the configured TICKET_PATTERN.
get_config
See current settings and their resolved values.
Prefix Examples
| Branch | Commit/PR Prefix |
|---|---|
feature/PROJ-123-add-login |
PROJ-123: |
task/update-readme |
Task: |
bug/fix-crash |
Bug: |
main |
(no prefix) |
PR Templates
PR Narrator automatically selects the best template for each repository through a resolution pipeline:
- Repo template — if a
PULL_REQUEST_TEMPLATE.mdexists in the repo (.github/, root, ordocs/), it's parsed into sections - Explicit preset — if
PR_TEMPLATE_PRESETis set ortemplatePresetis passed to a tool - Auto-detected domain — the repo's file tree is scanned and scored to detect its domain
- Default — a universal 6-section template
This means switching between repos (iOS app, Express API, Terraform infra) automatically uses the right template with zero configuration.
Domain Auto-Detection
PR Narrator scans the top 3 levels of the repo file tree and scores files against domain signal patterns. The domain with the highest score wins, as long as it reaches a minimum threshold.
| Domain | Key Signals | Sections Added |
|---|---|---|
| mobile | .swift, .kt, .xcodeproj, AndroidManifest.xml |
Screenshots, Device Testing, Accessibility |
| frontend | .tsx, .vue, next.config, vite.config |
Screenshots / Visual Changes, Browser Compatibility, Accessibility |
| backend | .go, .rs, migrations/, prisma/schema |
API Changes, Database / Migration, Breaking Changes |
| devops | .tf, helm/, k8s/, Dockerfile |
Infrastructure Impact, Affected Environments, Rollback Plan |
| security | .snyk, tfsec, trivy |
Security Impact, Threat Model Changes |
| ml | .ipynb, model/, training/, dvc.yaml |
Model Changes, Dataset Changes, Metrics / Evaluation |
Available Presets
| Preset | Sections | Best For |
|---|---|---|
default |
6 | General-purpose repos |
minimal |
2 | Quick PRs (Purpose + Test Plan) |
detailed |
10 | Thorough reviews with screenshots, breaking changes, deployment notes |
mobile |
8 | iOS and Android apps |
frontend |
8 | Web apps (React, Vue, Svelte, etc.) |
backend |
8 | APIs and services |
devops |
8 | Infrastructure and CI/CD |
security |
7 | Security-focused changes |
ml |
8 | Machine learning and data science |
Conditional Sections
Sections can appear or hide based on context:
has_tickets— Ticket section only appears when tickets are found in the branch name or commitsfile_pattern— Screenshots section only appears when UI files are changed; Database section only when migration files are changedcommit_count_gt— Changes (commit list) section only appears when there's more than 1 commit
Repo Template Detection
If your repo has a PULL_REQUEST_TEMPLATE.md, PR Narrator will find and parse it automatically. Supported locations:
.github/pull_request_template.md.github/PULL_REQUEST_TEMPLATE/(picksdefault.mdfirst)pull_request_template.md(repo root)docs/pull_request_template.md
File names are matched case-insensitively. Both .md and .txt extensions are supported.
Set PR_DETECT_REPO_TEMPLATE=false to skip repo template detection and use presets or auto-detection instead.
Security
This MCP server is read-only and local-only (stdio transport). It never modifies your git repository, makes network requests, or handles authentication tokens.
Things to be aware of:
- Diffs may contain secrets. Staged changes and branch diffs are sent to the AI for analysis. If your commits contain API keys or passwords, these will be visible to the model. Use git-secrets or gitleaks to prevent committing sensitive data.
- Commit messages are untrusted input. Git commit messages from collaborators are passed to the AI. Adversarial commit messages could theoretically attempt prompt injection. The read-only nature of this MCP limits impact.
- Regex patterns are validated. The
TICKET_PATTERNenv var is checked for ReDoS safety (catastrophic backtracking, length limits) before use.
For full details, see SECURITY.md.
Development
git clone https://github.com/mhaviv/pr-narrator-mcp.git
cd pr-narrator-mcp
npm install
npm run build
npm test
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。