Open source · Model Context Protocol · AI Dev Family

The mk-* family.
AI dev, decomposed.

AI 大師系列 · MK FAMILY

An opinionated family of MCP servers for the full AI-driven dev pipeline. Each member owns one stage — ideas ranked, specs produced, tests generated, coverage tracked, suite coached, trend self-reinforced over time. Compose them in your Claude / Cursor / Codex / Gemini config and let the AI drive.

Meet the family → View on GitHub
Works with any MCP-compatible client
Claude Desktop Cursor Codex CLI Gemini CLI Cline Zed
The loop

Plan → Specs → Code → Tests → Coverage → Coach

Three MCPs, one segment each, chained by the AI client. The middle 'Code' step deliberately lives in your IDE (Claude Code / Cursor / Copilot) — the family doesn't rewrite that layer. Read it as TDD: tests are the executable form of the spec, the IDE flips them red → green, then the coach loops you back to plan-master for the next idea.

Plan
plan-master
Linear / JIRA / Notion / Markdown · RICE ranking + senior-PM analysis + spec-draft bridge
Specs
spec-master
Linear / JIRA / Notion / Figma / GitHub Issues / Markdown
Code
your IDE
Claude Code · Cursor · Copilot — AI writes app code against red tests. The family deliberately doesn't touch this layer.
Tests
qa-master
pytest / Jest / Cypress / Go test / Maestro · web + mobile
Coverage
spec-master
spec ↔ test matrix · drift detection · auto-link via @spec tags
Coach
both
optimization plan · spec-quality grader · MCP usage telemetry
Current members

Three MCPs. One pipeline.

mk-plan-master prioritizes the ideas. mk-spec-master turns specs into scenarios and tracks coverage. mk-qa-master runs the tests. Together they form the idea → plan → spec → test → coverage → coach loop — composable in any MCP client, no MCP-to-MCP RPC required.

Execution side of the loop

MK QA Master · AI 測試大師

alpha · v0.6.1 16 tools

Run + analyze + advise — web (pytest / Jest / Cypress / Go), mobile (Maestro), and API (Schemathesis / Newman).

  • 7 runners across web + mobile + API — switch with one QA_RUNNER env var
  • analyze_url / analyze_screen — real DOM selectors + live mobile hierarchy, no hallucinated classes
  • generate_test — runnable pytest .py or Maestro .yaml; Schemathesis auto-fuzzes any OpenAPI schema
  • get_optimization_plan — classifies broken vs flaky vs slow-regression with evidence from run history
Deep dive → GitHub PyPI
Traceability side of the loop

MK Spec Master · AI 規格大師

alpha · v0.4.0 18 tools

Specs → scenarios → tests, with bidirectional spec ↔ test traceability.

  • 6 spec sources: Markdown, GitHub Issues, Linear, JIRA, Notion, Figma
  • Spec-quality coach: vague-language + implementation-leak + unclear-role heuristics
  • Coverage matrix + drift report (ac_hash-based)
  • auto_link_tests — scan @spec: tags from Python / JS / TS / Go test files
  • v0.4 self-reinforcement: history trend, chronic-drift signature, tool-usage telemetry
Deep dive → GitHub PyPI
Planning side of the loop

MK Plan Master · AI 規劃大師

alpha · v0.1.0 15 tools

Ideas in, prioritized plans out. Spec drafts that hand straight to mk-spec-master.

  • 4 adapters: markdown_local + Linear + JIRA + Notion
  • analyze_initiative — senior-PM analysis SOP (default / lite / lean_canvas frameworks)
  • generate_spec_draft hands off to mk-spec-master.parse_spec — verified end-to-end
  • RICE + Impact-Effort scoring with file-based decision index
  • Self-reinforcement built in: history + decision signature + telemetry
Deep dive → GitHub PyPI
Coming soon

Two more members in design

Same coach-layer pattern, different domains. Designed for the same MCP client config so adding the next layer is one block of JSON.

In design

MK Perf Master

in design

Performance audit MCP — Lighthouse / Core Web Vitals / bundle analyzer behind one coach.

In design

MK A11y Master

in design

Accessibility audit MCP — axe-core / Pa11y / WCAG 2.2 with prioritized remediation.

Install

Drop all three into your MCP client config

One JSON block. The AI client chains them automatically: idea → plan → spec → test.

{
  "mcpServers": {
    "mk-plan-master": {
      "command": "uvx",
      "args": ["mk-plan-master"],
      "env": {
        "PLAN_SOURCE": "markdown_local",
        "PLAN_PROJECT_ROOT": "/path/to/your/project"
      }
    },
    "mk-spec-master": {
      "command": "uvx",
      "args": ["mk-spec-master"],
      "env": {
        "SPEC_SOURCE": "markdown_local",
        "SPEC_PROJECT_ROOT": "/path/to/your/project"
      }
    },
    "mk-qa-master": {
      "command": "uvx",
      "args": ["mk-qa-master"],
      "env": {
        "QA_RUNNER": "pytest",
        "QA_PROJECT_ROOT": "/path/to/your/project"
      }
    }
  }
}