io.github.grahamnotgrant/blacksmith
编码与调试by grahamnotgrant
MCP server for Blacksmith CI - query runs, analyze test failures, detect flaky tests.
什么是 io.github.grahamnotgrant/blacksmith?
MCP server for Blacksmith CI - query runs, analyze test failures, detect flaky tests.
README
Blacksmith MCP
An MCP server that connects Claude to your Blacksmith CI data. Query workflow runs, analyze test failures, detect flaky tests, and monitor usage—all through natural conversation.
Why?
Debugging CI failures usually means clicking through dashboards, copying run IDs, and piecing together information across multiple pages. With this MCP, you can just ask:
- "Why did the last CI run fail?"
- "Which tests are flaky this week?"
- "Compare test failures between main and my PR"
- "What's using the most cache storage?"
Claude handles the API calls and gives you actionable insights.
Quick Start
Zero-config if you're logged into Blacksmith in Chrome:
# Add to Claude Code
claude mcp add blacksmith -- npx blacksmith-mcp
# Set your org (run once)
export BLACKSMITH_ORG="your-org-name"
The MCP automatically extracts your session from Chrome cookies. No manual token copying needed.
Installation
Option 1: Claude Code CLI
claude mcp add blacksmith -- npx blacksmith-mcp
Option 2: Project Configuration
Add to your .mcp.json:
{
"mcpServers": {
"blacksmith": {
"type": "stdio",
"command": "npx",
"args": ["blacksmith-mcp"],
"env": {
"BLACKSMITH_ORG": "your-org-name"
}
}
}
}
Option 3: Global Install
npm install -g blacksmith-mcp
Configuration
Authentication
Automatic (recommended): Log into app.blacksmith.sh in Chrome. The MCP extracts your session cookie automatically.
Manual: Set BLACKSMITH_SESSION_COOKIE environment variable with your session cookie value.
Environment Variables
| Variable | Required | Description |
|---|---|---|
BLACKSMITH_ORG | Yes | Your Blacksmith organization name |
BLACKSMITH_SESSION_COOKIE | No | Session cookie (auto-extracted from Chrome if not set) |
Available Tools
Workflow Runs
| Tool | Description |
|---|---|
list_runs | List workflow runs with filters (status, branch, workflow, actor, PR) |
get_run | Get run details including all jobs |
list_jobs | List jobs for a workflow run |
get_job | Get job details (steps, timing, runner info) |
get_job_logs | Get raw log output for a job |
Test Analytics
| Tool | Description |
|---|---|
get_job_tests | Get all test results for a job |
get_failed_tests | Get failed tests with full error messages |
get_failures_by_pattern | Group failures by error pattern (e.g., "Cannot read properties") |
compare_test_runs | Compare failures between two runs (find regressions) |
get_flaky_tests | Detect tests that fail intermittently |
get_slow_tests | Find tests exceeding duration threshold |
get_test_history | Track a specific test's pass/fail history |
get_trends | Analyze trends: duration, failure rate, test count |
Usage & Billing
| Tool | Description |
|---|---|
get_current_usage | Current core usage (active vs max) |
get_invoice_amount | Current billing period amount |
get_usage_summary | Billable minutes vs free tier |
get_cache_stats | Cache size, entries by repository |
get_cache_entries | Detailed cache entries for a repo |
Organization
| Tool | Description |
|---|---|
list_orgs | List accessible organizations |
get_org_status | Org status (personal, onboarded, region) |
search_logs | Search logs across all jobs |
Example Conversations
Debug a Failed Run
You: Why did the last CI run fail?
Claude: Uses
list_runs(status="failure", limit=1)thenget_failed_testsThe last run failed with 34 test failures. The main issues are:
- Connection timeout (12 tests) - Tests in
api/are timing out connecting to the test database- Undefined property (8 tests) -
user.settings.themeis undefined in the new dark mode code- Assertion failures (14 tests) - Snapshot mismatches in UI components
Find Flaky Tests
You: Which tests are flaky?
Claude: Uses
get_flaky_tests(days=7, min_runs=5)Found 3 flaky tests:
Test Flake Rate Pattern should handle concurrent requests40% (4/10 runs) Fails on CI, passes locally renders loading state20% (2/10 runs) Race condition with setTimeout validates auth token30% (3/10 runs) Intermittent network timeout
Compare Branches
You: Did my PR introduce new test failures?
Claude: Uses
compare_test_runswith PR run vs base branchYour PR introduced 2 new failures:
UserProfile.test.tsx: "Cannot read property 'avatar' of undefined"api/users.test.ts: Expected 200, got 4045 previously failing tests now pass (nice!).
Development
# Install dependencies
pnpm install
# Build
pnpm build
# Development mode (watch)
pnpm dev
# Test with MCP Inspector
npx @modelcontextprotocol/inspector node dist/index.js
Troubleshooting
Session Expired
If you see SESSION_EXPIRED, your Blacksmith session has expired. Simply log back into app.blacksmith.sh in Chrome and retry.
Cookie Extraction Failed
The automatic cookie extraction requires:
- macOS with Chrome installed
- Being logged into Blacksmith in Chrome
- Chrome not running with a locked profile
If it fails, set BLACKSMITH_SESSION_COOKIE manually.
No Organization Set
Run list_orgs to see available organizations, then set BLACKSMITH_ORG to your org name.
API Notes
This MCP uses Blacksmith's internal web API, which is undocumented. The API was reverse-engineered from the Blacksmith web app and may change without notice.
License
MIT
Contributing
Contributions welcome! Please open an issue first to discuss proposed changes.
常见问题
io.github.grahamnotgrant/blacksmith 是什么?
MCP server for Blacksmith CI - query runs, analyze test failures, detect flaky tests.
相关 Skills
网页构建器
by anthropics
面向复杂 claude.ai HTML artifact 开发,快速初始化 React + Tailwind CSS + shadcn/ui 项目并打包为单文件 HTML,适合需要状态管理、路由或多组件交互的页面。
✎ 在 claude.ai 里做复杂网页 Artifact 很省心,多组件、状态和路由都能顺手搭起来,React、Tailwind 与 shadcn/ui 组合效率高、成品也更精致。
前端设计
by anthropics
面向组件、页面、海报和 Web 应用开发,按鲜明视觉方向生成可直接落地的前端代码与高质感 UI,适合做 landing page、Dashboard 或美化现有界面,避开千篇一律的 AI 审美。
✎ 想把页面做得既能上线又有设计感,就用前端设计:组件到整站都能产出,难得的是能避开千篇一律的 AI 味。
网页应用测试
by anthropics
用 Playwright 为本地 Web 应用编写自动化测试,支持启动开发服务器、校验前端交互、排查 UI 异常、抓取截图与浏览器日志,适合调试动态页面和回归验证。
✎ 借助 Playwright 一站式验证本地 Web 应用前端功能,调 UI 时还能同步查看日志和截图,定位问题更快。
相关 MCP Server
GitHub
编辑精选by GitHub
GitHub 是 MCP 官方参考服务器,让 Claude 直接读写你的代码仓库和 Issues。
✎ 这个参考服务器解决了开发者想让 AI 安全访问 GitHub 数据的问题,适合需要自动化代码审查或 Issue 管理的团队。但注意它只是参考实现,生产环境得自己加固安全。
Context7 文档查询
编辑精选by Context7
Context7 是实时拉取最新文档和代码示例的智能助手,让你告别过时资料。
✎ 它能解决开发者查找文档时信息滞后的问题,特别适合快速上手新库或跟进更新。不过,依赖外部源可能导致偶尔的数据延迟,建议结合官方文档使用。
by tldraw
tldraw 是让 AI 助手直接在无限画布上绘图和协作的 MCP 服务器。
✎ 这解决了 AI 只能输出文本、无法视觉化协作的痛点——想象让 Claude 帮你画流程图或白板讨论。最适合需要快速原型设计或头脑风暴的开发者。不过,目前它只是个基础连接器,你得自己搭建画布应用才能发挥全部潜力。