What do I need to run Blacksmith?
You need a Blacksmith organization name, set as the `BLACKSMITH_ORG` environment variable. The server is run via `npx blacksmith-mcp`.
B
Blacksmith MCP Server
@grahamnotgrant
基本信息
关于 Blacksmith MCP Server
Connect Claude to your Blacksmith CI data. Query workflow runs, analyze test failures, detect flaky tests, compare test regressions between branches, and monitor usage—all through natural conversation. Features 21 tools including test analytics, cache stats, and log search. Zero-
Blacksmith MCP Server 配置
使用下面的配置,将此服务器添加到你的 MCP 客户端。
{
"mcpServers": {
"blacksmith": {
"command": "npx",
"args": [
"blacksmith-mcp"
],
"env": {
"BLACKSMITH_ORG": "<YOUR_ORG_NAME>"
}
}
}
}Blacksmith MCP Server 工具
21List all Blacksmith organizations accessible to your account. Use this first to see available orgs.
Get the status of the current organization (personal org, onboarding, runner region).
List workflow runs with filtering. Filter by status (success/failure/cancelled/skipped/in_progress), branch, workflow name, actor, or PR number. Example: list_runs(status="failure") to find failed runs.
Get details of a specific workflow run by ID. Includes list of jobs.
List all jobs for a specific workflow run. Use this to get job IDs for get_job, get_job_logs, and get_job_tests.
Get details of a specific job including steps, runner info, and timing.
Get the logs for a specific job. Returns raw log output.
Get test results for a job. Optionally filter by status (pass/fail/skip).
Get failed tests for a job with full error details. Use error_lines param to control stack trace length. Returns all failures by default (no limit).
Group failed tests by error pattern (e.g., "is not a function", "Cannot read properties"). Shows count, affected suites/files, and sample error for each pattern. Best for quickly identifying root causes.
Compare test failures between two runs to identify regressions. Shows new failures, fixed tests, and persistent failures. If base_run_id not provided, compares against most recent prior run.
Detect flaky tests by analyzing pass/fail patterns across recent runs. Returns tests that fail intermittently (e.g., "failed 3 of 10 runs"). Killer feature for CI stability.
Find tests exceeding a duration threshold. Shows slowest tests, their percentage of total test time, and average duration stats.
Get the failure history for a specific test across recent runs. Shows when it passed/failed, on which branches, and error messages for failures.
Track metrics over time: duration (are tests getting slower?), failure_rate (are tests getting flakier?), test_count (are we adding tests?). Returns trend analysis with data points.
Get current core usage snapshot (active cores vs max cores).
Get the current billing period invoice amount.
Get usage summary showing billable minutes vs free tier allowance. Shows remaining free minutes and overage.
Get Blacksmith cache statistics: total size, hit rate, entries by repository. Shows how effectively caching is being used.
Get detailed cache entries for a repository. Shows cache keys, sizes, scopes (branches), and last hit times. Useful for debugging cache issues.
Search logs across all jobs. Filter by query (e.g., "error", "timeout"), log level (INFO/WARN/ERROR/DEBUG), and time range. Great for finding issues across runs.
概览
What is Blacksmith?
Blacksmith is an MCP server that connects Claude to GitHub Actions CI/CD analytics and monitoring. It enables developers to debug CI failures, detect flaky tests, and monitor workflow runs directly through natural language queries.
How to use Blacksmith?
Install via Claude: claude mcp add blacksmith -- npx blacksmith-mcp, then set the environment variable BLACKSMITH_ORG=your-org-name. After setup, you can ask Claude questions such as "Why did the last CI run fail?" or "Which tests are flaky this week?"
Key features of Blacksmith
- Flaky test detection and trend analysis
- Test failure comparison between branches
- Error pattern grouping by message
- Slow test finder with duration thresholds
- Workflow run listing and filtering by status, branch, actor, or PR
- Job detail inspection including steps and timing
- Log search across all jobs
- Core usage and billing monitoring
- Cache statistics per repository
- Zero-config authentication via Chrome cookies
Use cases of Blacksmith
- Determine why the latest CI run failed and which tests are affected
- Identify flaky tests that intermittently fail over time
- Compare test failure rates between branches to catch regressions
- Monitor workflow performance and detect slow jobs or steps
- Track billing and resource usage for GitHub Actions
常见问题
How does authentication work?
Blacksmith automatically extracts your session from Chrome cookies, eliminating the need to copy tokens manually.
What can I ask Claude to do with Blacksmith?
You can ask about CI pipeline status, flaky tests, test comparisons between branches, detailed job logs, and usage/cache statistics.
Does Blacksmith require any additional setup beyond the environment variable?
No other setup is mentioned. The zero-config auth uses existing Chrome cookies, so no token generation is needed.
开发工具 分类下的更多 MCP 服务器
Launchdog
Devin StarksLaunch your product from your AI agent. Connect Claude Code, Codex, Cursor or ChatGPT to Launchdog over MCP and your agent drafts the listing, picks a launch week and books it. 50 launches a week, ranked only by human vo
offprompt
offpromptMCP server that lets Claude Code, Cursor or Codex ask you for API keys on a local page and write them to .env. The agent only ever sees key names.
2
LocalCan
LocalCanGives AI agents public URLs (tunnels) for localhost, live HTTP traffic inspection, snapshot publishing, and access control. Part of LocalCan, the ngrok alternative for Mac, Windows and Linux. Free plan.
82
LastPing
LastPingKnow when your AI agents, cron jobs and CI pipelines fail, stall or go quiet. Your agent sets up monitors from a plain description, tells you what is broken and why, reads run history and traces, and writes notes on inci
OpenMark
OpenMark AIOpenMark is a remote MCP server available at https://openmark.ai. Benchmark your own prompt on 100+ models with real API calls.
评论