IDE 및 웹 애플리케이션 전반에서 구현 계획, 코드 리뷰, 자동화된 테스트를 포함한 기술 워크플로우를 AI 에이전트가 실행할 수 있도록 지원합니다.
다른 이름 bronz3beard/ai.tech-lead-stack, ai-tech-lead-stack, tech-lead-stack
- 스타
- 3
- Tools
- —
- 다운로드, 30일
- —
- 최신 버전
- —
- 마지막 커밋
- 2026년 11월 14일
- 라이선스
- MIT
Tech Lead Stack MCP란?
IDE 및 웹 애플리케이션 전반에서 구현 계획, 코드 리뷰, 자동화된 테스트를 포함한 기술 워크플로우를 AI 에이전트가 실행할 수 있도록 지원합니다.
프로젝트 README에서 생성된 요약
Tech Lead Stack MCP server 정보
기술 리드와 개발자는 이 에이전트 불가지론적(agent-agnostic) 스택을 통해 구현 계획, 코드 리뷰, 자동화된 테스트를 포함한 복잡한 엔지니어링 워크플로우를 자동화할 수 있습니다. 초기 의도 요약부터 기술 사양, 구현 계획, 최종 릴리스에 이르기까지 구조화된 파이프라인을 제공합니다. 이 시스템은 Claude, Gemini, GPT와 같은 다양한 AI 에이전트에서 작동하도록 설계되었으며, IDE에 직접 통합하거나 웹 대시보드를 통해 사용할 수 있습니다.
사용자는 GitHub 저장소에 연결되는 웹 애플리케이션을 사용하거나, apply_patch 도구를 사용하여 로컬 파일을 수정하는 IDE 코딩 에이전트를 통해 시스템과 상호작용할 수 있습니다. 설정을 위해서는 GitHub 계정 연결과 선택한 AI 제공업체의 API 키가 필요합니다. 고급 기능을 탐색할 때, 시스템은 E2B 샌드박스를 활용하여 프로젝트 의존성이 설치된 개발 서버를 부팅함으로써 라이브 환경에서 변경 사항을 미리 볼 수 있습니다. 이를 통해 코드를 저장소에 커밋하기 전에 격리된 환경에서 구현 아이디어를 테스트할 수 있습니다.
활용 사례
- 1안전한 기능 프로토타이핑개발자는 메인 코드베이스에 영향을 주지 않고 샌드박스 환경에서 새로운 기능을 미리 볼 수 있습니다.
- 2반복적인 계획 정교화reflexion 엔진이 기술 계획을 초안 작성, 비판 및 수정하여 구현 전 품질을 보장합니다.
프로젝트의 README를 바탕으로 작성되었습니다. 모든 활용 사례는 해당 내용의 인용문과 대조하여 확인합니다.
Tech Lead Stack MCP 설치 방법
형식 확인: 2026년 11월 10일Cursor에 추가Goose에 추가Claude Code에 Tech Lead Stack MCP 설치
$ claude mcp add ai-tech-lead-stack -- npx -y tech-lead-stackCursor에 Tech Lead Stack MCP 설치
~/.cursor/mcp.json
{
"mcpServers": {
"ai-tech-lead-stack": {
"command": "npx",
"args": [
"-y",
"tech-lead-stack"
]
}
}
}VS Code에 Tech Lead Stack MCP 설치
.vscode/mcp.json
{
"servers": {
"ai-tech-lead-stack": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"tech-lead-stack"
]
}
}
}VS Code CLI에 Tech Lead Stack MCP 설치
$ code --add-mcp '{"name":"ai-tech-lead-stack","type":"stdio","command":"npx","args":["-y","tech-lead-stack"]}'Claude Desktop에 Tech Lead Stack MCP 설치
claude_desktop_config.json
{
"mcpServers": {
"ai-tech-lead-stack": {
"command": "npx",
"args": [
"-y",
"tech-lead-stack"
]
}
}
}Codex에 Tech Lead Stack MCP 설치
$ codex mcp add ai-tech-lead-stack -- npx -y tech-lead-stackCodex config.toml에 Tech Lead Stack MCP 설치
~/.codex/config.toml
[mcp_servers.ai-tech-lead-stack] command = "npx" args = ["-y", "tech-lead-stack"]
Gemini CLI에 Tech Lead Stack MCP 설치
$ gemini mcp add -s user ai-tech-lead-stack npx -- -y tech-lead-stackGemini settings.json에 Tech Lead Stack MCP 설치
~/.gemini/settings.json
{
"mcpServers": {
"ai-tech-lead-stack": {
"command": "npx",
"args": [
"-y",
"tech-lead-stack"
]
}
}
}자동 검사
2026년 11월 14일에 실행됨전체 보고서 →자동 검사를 통과했습니다. 이는 보안 감사가 아닙니다. 코드를 직접 읽고 서버에 필요한 최소한의 권한만 부여하십시오. 검사 방식
- 소스 저장소 연결 및 접근 가능bronz3beard/ai.tech-lead-stack is public and reachablegithub.com/bronz3beard/ai.tech-lead-stack ↗통과
- 패키지가 이 저장소를 소스로 명시tech-lead-stack names bronz3beard/ai.tech-lead-stack as sourcenpmjs.com/package/tech-lead-stack ↗통과
- 빌드 출처(Sigstore 증명)Sigstore attestation for tech-lead-stacknpmjs.com/package/tech-lead-stack/v/2.3.0#provenance ↗통과
- 오픈 소스 라이선스 (OSI)MITspdx.org/licenses/MIT.html ↗통과
- 최근 1년 내 커밋 또는 릴리스Last commit or release 2026-11-07github.com/bronz3beard/ai.tech-lead-stack ↗통과
- OpenSSF ScorecardOpenSSF Scorecard 8.0 / 10; weakest: Code-Review 0, Vulnerabilities 0, Contributors 3scorecard.dev/viewer/?uri=github.com/bronz3beard/ai.tech-lea ↗통과
- 패키지 이름이 인기 패키지와 유사하지 않음No popular package one edit away from tech-lead-stack통과
- 패키지 및 해당 종속성에 중요한 보안 권고 없음검사되지 않음
- 모델에 대한 숨겨진 지침 없음0 tool descriptions and the README read; no override phrases, credential paths or hidden characters통과
- 파괴적인 도구에 라벨 표시됨검사되지 않음
- 원격 엔드포인트가 MCP 핸드셰이크에 응답검사되지 않음
README
GitHub에서 보기The Lead Stack: Agent-Agnostic Workflows
intent-brief → spec → plan → diff → review-report → release
A high-performance repository of "Skills" and RTK-powered tools designed for Tech Leads. These workflows are Agent-Agnostic, allowing any LLM agent (Gemini, Claude, GPT) to assist with implementation planning, code review, and automated testing.
Live Web App: https://ai-tech-lead-stack.vercel.app
Setting this up with an AI assistant? There's a prompt written for that: Set this up with an AI assistant. It checks your computer, asks what you want help with, sets everything up, and answers your questions afterwards.
Contents
- How it fits together
- Commands Quick Reference
- Which tier am I on?
- Requirements
- Set this up with an AI assistant
- 🚀 Quick Start
- Supported Editors & Agents
- 🧹 Resetting a Project
- Documentation: setup guides, the skill and workflow catalogues, methodology and architecture
- Peripherals & Sibling Apps
- Branching Strategy
- Support, Contributing & Security
How it fits together
There are two ways in: the web dashboard and your IDE's coding agent. Both use
the same shared core (packages/core). Both can also change code:
- The web app mostly reads your linked GitHub repo, but a few routes commit
to its
feat/*anddiscovery/*branches or trigger its GitHub Actions (close-up 4). - In the IDE, your coding agent edits files, and the MCP server's
apply_patchtool can write them too (close-up 7).
The overview comes first. It is followed by the web app flow, then the IDE flow, and finally the telemetry both flows share. Colours are the same in every diagram: purple for people, blue for the dashboard, green for the shared core and MCP server, and pink for storage and outside services. Dashed lines are optional or conditional calls. Thick arrows are writes to code or to a Git repo.
flowchart LR
dev(("Developer<br/>in browser"))
agent(("IDE coding agent"))
dash["Dashboard app<br/>apps/dashboard"]
subgraph core["Shared core · packages/core"]
mcp["MCP server"]
libs["Reflexion engine · skills<br/>model resolver · telemetry"]
end
repo["Your project repo<br/>on GitHub"]
code[("Files on your machine<br/>(MCP working directory)")]
ext[("Postgres · AI providers<br/>E2B · ClickUp · Figma · Discord · Ollama")]
dev -->|"uses"| dash
agent -->|"calls MCP tools"| mcp
agent ==>|"edits"| code
mcp ==>|"reads · apply_patch writes"| code
dash -->|"uses"| libs
mcp -->|"uses"| libs
dash ==>|"reads · commits to feat/* and discovery/*<br/>branches · triggers Actions"| repo
dash -->|"app data · models · sandboxes · webhooks"| ext
libs -->|"app data · models · telemetry"| ext
mcp -->|"code_search embeddings"| ext
classDef actor fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef app fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef coreNode fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class dev,agent actor
class dash app
class mcp,libs coreNode
class repo,code,ext extNode
Web app flow
Everything the dashboard (apps/dashboard) does. You sign in with GitHub, and
routes that touch GitHub use your GitHub token on the project's linked repo.
Model calls use the API keys you saved in Settings, and fall back to the
server's env keys.
1. Chat and feature discovery
Chat streams replies from your chosen model and saves the conversation. When the
project has a linked repo, it reads code from that repo on GitHub. Otherwise its
tools read skills and files from the dashboard server's own toolbox checkout.
Slash-command workflows always come from the toolbox. The ClickUp and Figma
tools only read, and only work when the project has those keys saved. Feature
discovery streams a planning conversation; its one tool, write_to_sandbox,
runs in your browser (close-up 3). Both routes also read your user and project
records.
flowchart LR
discovery["Feature discovery<br/>api/orchestrator/discovery"]
chat["Chat<br/>api/chat"]
db[("Postgres")]
ai["AI model providers"]
tools["ClickUp · Figma<br/>read-only"]
github["Your project repo<br/>on GitHub"]
toolbox[("Toolbox skills<br/>and workflows")]
core["Telemetry<br/>(see close-up 9)"]
discovery -->|"streams replies"| ai
chat -->|"saves chats and messages"| db
chat -->|"streams replies"| ai
chat -.->|"reads tasks and designs"| tools
chat -.->|"reads code (repo linked)"| github
chat -->|"slash workflows · skills<br/>when no repo is linked"| toolbox
chat -->|"records events"| core
click chat "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/chat/route.ts"
click discovery "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/orchestrator/discovery/route.ts"
classDef actor fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef app fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef coreNode fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class chat,discovery app
class core coreNode
class tools,github,toolbox,ai,db extNode
2. Reflexion from the web app
The Reflexion route reads package.json and tsconfig.json from your linked
repo for context, then runs the reflexion engine. The engine drafts a plan,
critiques it and revises it, saving progress to the ReflexionRun table after
each step. Every step is also recorded as telemetry. You get back a plan and an
IDE prompt to apply yourself. The resume route continues a paused run the same
way, without reading GitHub.
flowchart LR
dev(("Developer<br/>in browser"))
reflexapi["Reflexion route<br/>api/orchestrator/reflexion"]
github["Your project repo<br/>on GitHub"]
engine["Reflexion engine<br/>engine.ts"]
runner["Model runner<br/>providers-user"]
ai["AI model providers"]
dbstore["DbStateStore"]
db[("Postgres<br/>ReflexionRun")]
telemetry["Telemetry<br/>(see close-up 9)"]
dev -->|"starts a run"| reflexapi
reflexapi -.->|"reads package.json<br/>and tsconfig.json"| github
reflexapi -->|"runs loop"| engine
engine -->|"drafts and critiques"| runner
runner -->|"your saved keys,<br/>else env keys"| ai
engine -->|"saves progress"| dbstore
dbstore --> db
reflexapi -->|"records each step"| telemetry
reflexapi -->|"plan + IDE prompt"| dev
click reflexapi "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/orchestrator/reflexion/route.ts"
click engine "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/lib/ai/reflexion/engine.ts"
click runner "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/lib/ai/reflexion/providers-user.ts"
click dbstore "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/lib/ai/reflexion/db-state-store.ts"
classDef actor fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef app fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef coreNode fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class dev actor
class reflexapi app
class engine,runner,dbstore,telemetry coreNode
class github,ai,db extNode
3. Sandbox preview
Feature discovery can preview changes in an E2B sandbox, using the E2B key you saved in Settings. Your browser first fetches a zip of the linked repo's default branch, together with the project's decrypted env vars. It then asks the server to boot a sandbox, which uploads the project, installs dependencies and starts a dev server. File writes from the discovery AI go into that sandbox only. Nothing is copied back to GitHub automatically.
flowchart LR
dev(("Developer<br/>in browser"))
bundle["Project bundle<br/>api/orchestrator/discovery/project-bundle"]
repo["Your project repo<br/>on GitHub"]
boot["Sandbox boot<br/>api/orchestrator/sandbox/boot"]
fileaction["Sandbox file action<br/>discovery/actions/e2b.ts"]
e2b["E2B sandbox<br/>(your E2B key)"]
dev -->|"1 · fetch project"| bundle
bundle -->|"downloads a zip of<br/>the default branch"| repo
dev -->|"2 · boot sandbox"| boot
boot -->|"uploads, installs,<br/>starts dev server"| e2b
dev -->|"3 · AI file writes"| fileaction
fileaction -->|"writes files"| e2b
click bundle "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/orchestrator/discovery/project-bundle/route.ts"
click boot "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/orchestrator/sandbox/boot/route.ts"
click fileaction "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/feature-development/discovery/actions/e2b.ts"
classDef actor fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef app fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class dev actor
class bundle,boot,fileaction app
class repo,e2b extNode
4. Writes to your GitHub repo
These routes change your linked repo on GitHub, using your GitHub token.
Feedback commits a screenshot to .github/assets/feedback/ on a branch,
comments on that branch's pull request and posts to Discord. Sync commits
whatever files it is sent; nothing in the dashboard calls it yet. Both refuse
any branch that isn't feat/* or discovery/*, and always block main,
master, prod, production and staging. Generate and Audit trigger the
ai-prompt-agent.yml and audit.yml GitHub Actions on your repo's main
branch. Those workflow files live in your repo, so what they change is defined
there, not here.
flowchart LR
feedback["Feedback<br/>api/feature-development/feedback"]
sync["Sync<br/>api/feature-development/sync"]
actions["Generate · Audit<br/>api/orchestrator/generate · audit"]
repo["Your project repo<br/>on GitHub"]
discord["Discord"]
feedback ==>|"commits snapshot PNG<br/>· comments on PR"| repo
feedback -->|"notifies"| discord
sync ==>|"commits files to<br/>feat/* or discovery/*"| repo
actions ==>|"triggers ai-prompt-agent.yml<br/>or audit.yml"| repo
click feedback "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/feature-development/feedback/route.ts"
click sync "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/feature-development/sync/route.ts"
click actions "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/orchestrator/generate/route.ts"
classDef app fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class feedback,sync,actions app
class repo,discord extNode
5. Settings, design review and the skill editor
Projects and Settings store app data in Postgres: projects, who can access them, your API keys (encrypted), and per-project model routing. Design review keeps review sessions in Postgres, reads Figma files through a proxy with the project's Figma key, and posts to the project's Discord webhook when a review is ready for the designer. Notify devs posts to a separate Discord webhook. The skill editor drafts new skills with your model. When you submit one, it opens a draft pull request on this toolbox's own repo, not on your project.
flowchart LR
settings["Projects · Settings<br/>api/projects · api/settings/*"]
design["Design review<br/>api/design-review"]
notify["Notify devs<br/>api/notify/devs"]
skilled["Skill editor<br/>api/skills/chat · skills/actions.ts"]
db[("Postgres")]
figma["Figma"]
discord["Discord"]
ai["AI model providers"]
tbrepo["Toolbox repo<br/>on GitHub"]
settings -->|"projects, access, encrypted<br/>API keys, model routing"| db
design -->|"review sessions"| db
design -.->|"figma-proxy reads files"| figma
design -->|"ready for designer"| discord
notify -->|"webhook"| discord
skilled -->|"drafts skills"| ai
skilled ==>|"opens a draft PR<br/>with the new skill"| tbrepo
click design "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/design-review/route.ts"
click notify "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/notify/devs/route.ts"
click skilled "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/apps/dashboard/src/app/api/skills/actions.ts"
classDef app fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class settings,design,notify,skilled app
class db,figma,discord,ai,tbrepo extNode
IDE flow
Everything your coding agent does through the MCP server, which runs on your
machine. Every project path the server uses is relative to the directory it
was started in. The default install starts it with npm --prefix <toolbox>,
so that directory is the toolbox checkout unless your editor or a proxy starts
it somewhere else.
6. Skills, knowledge items and pipelines
Skills, workflows, policies and the skill graph are read from the toolbox
checkout. If the server finds a client project above its working directory, it
also reads that project's .ai overrides. Before any get_* call, the server
checks the hook rules in .ai/hooks and can refuse the call. Knowledge Items
(KIs) are stored in your home directory, in ~/.gemini/antigravity/knowledge.
verify_mission_alignment writes .ai/.mission-alignment.json into the toolbox
checkout. get_* and plan_pipeline calls are recorded as telemetry.
flowchart LR
agent(("IDE coding agent"))
mcp["MCP server<br/>mcp-server/index.ts"]
hooks[(".ai/hooks<br/>(working directory)")]
toolbox[("Toolbox files<br/>.ai/skills · workflows · policies")]
project[("Project .ai overrides<br/>skills.graph.json · policies")]
kis[("Knowledge items<br/>~/.gemini/antigravity/knowledge")]
telemetry["Telemetry<br/>(see close-up 9)"]
agent -->|"skill, pipeline,<br/>KI and alignment tools"| mcp
mcp -->|"get_* calls checked first,<br/>can refuse"| hooks
mcp -->|"reads skills, graph, policies<br/>· writes .mission-alignment.json"| toolbox
mcp -.->|"reads, if a client project is found"| project
mcp -->|"reads and writes KIs"| kis
mcp -->|"get_* and plan_pipeline<br/>record events"| telemetry
click mcp "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/mcp-server/index.ts"
classDef actor fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef coreNode fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class agent actor
class mcp,telemetry coreNode
class hooks,toolbox,project,kis extNode
7. Code tools
repo_map and code_search read a prebuilt index in .tls-index, which you
create with npm run index:build in packages/core. code_search also calls a
local Ollama (OLLAMA_URL, default localhost:11434) to embed your query.
read_region reads a line range from a file. apply_patch writes a file after
checking .ai/hooks for protected paths. None of these tools restrict paths to
the working directory.
flowchart LR
agent(("IDE coding agent"))
mcp["MCP server<br/>handlers/codebase.ts"]
index[(".tls-index<br/>built by index:build")]
ollama["Ollama<br/>localhost:11434"]
hooks[(".ai/hooks<br/>protected paths")]
files[("Files in the MCP<br/>working directory")]
agent -->|"repo_map · code_search<br/>read_region · apply_patch"| mcp
mcp -->|"repo_map, code_search read"| index
mcp -->|"code_search embeds query"| ollama
mcp -->|"apply_patch checks first"| hooks
mcp -->|"read_region reads"| files
mcp ==>|"apply_patch writes"| files
agent ==>|"edits"| files
click mcp "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/mcp-server/handlers/codebase.ts"
classDef actor fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef coreNode fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class agent actor
class mcp coreNode
class index,ollama,hooks,files extNode
8. Reflexion from the IDE
reflexion_loop starts a run in the background and returns a run id straight
away. Its _sub_max and _sub_pro variants only change the tier check. Before
starting, it looks you up in Postgres by your GitHub CLI or git email, along
with the project, so it can use the API keys and model routing you saved in the
web app. Env keys are the fallback, and env model variables such as
MODEL_PLANNER override saved routing. Progress is saved to
.reflexion-out/state.json, or to your OS temp directory when the server runs
inside another project. Poll reflexion_status for the result, and use
reflexion_resume to continue a paused run; resume skips the Postgres lookup
and uses env keys only. Each step is recorded as telemetry, and your coding
agent applies the final plan.
flowchart LR
agent(("IDE coding agent"))
mcp["MCP server<br/>handlers/reflexion.ts"]
db[("Postgres<br/>saved keys · model routing")]
engine["Reflexion engine<br/>engine.ts"]
runner["Model runner<br/>providers-env"]
ai["AI model providers"]
filestore["FileStateStore"]
files[(".reflexion-out/state.json<br/>or OS temp dir")]
telemetry["Telemetry<br/>(see close-up 9)"]
agent -->|"reflexion_loop · status · resume"| mcp
mcp -.->|"reflexion_loop only: looks up<br/>you (gh/git email) and project"| db
mcp -->|"runs loop in background"| engine
engine -->|"drafts and critiques"| runner
runner -->|"saved keys, else env keys<br/>(resume: env keys only)"| ai
engine -->|"saves progress"| filestore
filestore --> files
mcp -->|"records each step"| telemetry
mcp -->|"run id, then status and plan"| agent
click mcp "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/mcp-server/handlers/reflexion.ts"
click engine "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/lib/ai/reflexion/engine.ts"
click runner "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/lib/ai/reflexion/providers-env.ts"
click filestore "https://github.com/bronz3beard/ai.tech-lead-stack/blob/main/packages/core/src/lib/ai/reflexion/state-store.ts"
classDef actor fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef coreNode fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef extNode fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
class agent actor
class mcp,engine,runner,filestore,telemetry coreNode
class db,ai,files extNode
Shared by both flows
Tech Lead Stack MCP 질문
Tech Lead Stack MCP를 Claude Code에 추가하려면 어떻게 하나요?
터미널에서 claude mcp add ai-tech-lead-stack -- npx -y tech-lead-stack를 실행합니다. 모든 프로젝트에서 사용할 수 있게 하려면 --scope user를 추가합니다.
Tech Lead Stack MCP를 Cursor에 추가하려면 어떻게 하나요?
~/.cursor/mcp.json를 열고 mcpServers 아래에 서버를 추가합니다. 위 설치 섹션에 표시된 Cursor 구성을 사용합니다.
어떤 종류의 워크플로우를 지원하나요?
의도 요약(intent-brief), 사양 정의, 계획 수립, diff 생성, 리뷰 보고, 릴리스로 이어지는 연속적인 워크플로우를 지원합니다.