genie

MCP开发工具

提供方 automagik-dev · 包名 genie

开发工具 stdio · 本地进程 未标注 社区 / 第三方 Claude DesktopCursorClineClaude Code
查看接入方式

Wishes in, PRs out. CLI agent that interviews you, plans the work, dispatches parallel agents in isolated worktrees, and reviews code before you see i

一句话选型结论

按资料字段整理

genie 属于开发工具类 MCP 服务器,stdio · 本地进程,鉴权方式未标注,可直接接入 Claude Desktop、Cursor、Cline 等,有可直接执行的安装命令。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类开发工具 传输方式stdio · 本地进程 鉴权未标注 部署方式本地部署
GitHub 星标 280 仓库热度,无公开仓库时显示未标注
工具数 未标注 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

开发工具
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量未标注暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

genie ai-agents ai-developer-tools anthropic automation

支持客户端

适用场景

genieai-agentsai-developer-tools

安装与配置

stdio · 本地进程 未标注
客户端在本机拉起子进程,数据不出本机
npx -y genie

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    stdio 服务在本地起进程,需要先装好 Cross-platform,再按安装命令拉取依赖。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后随便调一个只读工具,确认能返回结果,再接入正式流程。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
代码与仓库操作

在对话里直接查代码、提 issue、走 PR,省掉来回切工具。

本地命令与脚本

把构建、测试、格式化这类重复命令交给模型执行。

接口联调与测试

让模型调你的接口、跑一遍用例,快速定位问题。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude Desktop、Cursor、Cline,换客户端不用重新研究配置。

先想清楚的地方

  • stdio 方式要求本地装好运行时(如 Node.js),换机器要重新装一遍。
  • 资料没标注工具数量,实际能力要装好后确认。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在开发工具场景里让模型直接调用外部能力的人。
  • 已经在用 Claude Desktop、Cursor 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 完全不想在本地装环境的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
genie当前 未标注 stdio · 本地进程 本地部署 有 2026-04-10
Git 838e2ec1 12 stdio · 本地进程 本地部署 有 2026-05-17
Markitdown MCP 2c653a5f 1 stdio · 本地进程 本地部署 有 2026-05-17
JavaGuide 未标注 stdio · 本地进程 本地部署 有 2026-04-11
awesome-claude-skills 未标注 stdio · 本地进程 本地部署 有 2026-04-26
Context7 831acf40 未标注 stdio · 本地进程 本地部署 有 2026-05-17

服务器档案

开发工具
字段来自项目仓库与公开索引,中文说明由本站标注
提供方automagik-dev上游标注的维护方 automagik-dev
分类开发工具本站按上游分类展示
连接方式stdio · 本地进程客户端在本机拉起子进程,数据不出本机
鉴权方式未标注远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时genie · Cross-platform本地安装时用到的包标识与运行时
部署类型本地部署上游给出的部署形态
最近核验2026-04-10上游最后核验时间

详细介绍

14.7 KB
来自项目仓库与上游目录的原文,未做删改

Code is a commodity. Harnesses get replaced. Context stays.

A time-lapsed replay of a real /wish run, host details hidden and agent launch line removed. Its figures are one labelled sample, from its own genie wish report. Full video

Give /wish one decided task. It checks that the task fits, builds it in its own worktree, runs your repository's own check, has an agent that did not write the code review the exact commit, and opens a pull request. Merging stays with you.```bash
curl -fsSL https://raw.githubusercontent.com/automagik-dev/genie/main/install.sh | bash


Claude Code runs it as `/wish`, Codex as `$wish`, and other agents by name. Docs · Skills · Discord

One `/wish` from a real run, shortened:```text
You:    /wish Delete dead code: src/lib/defaults.ts has no production callers
        (only a type import of AgentDefaults in src/lib/workspace.ts) and
        src/__tests__/dir-ls.test.ts reimplements a removed handler and asserts
        on its own copy. Move the AgentDefaults type into workspace.ts, delete
        defaults.ts, defaults.test.ts and dir-ls.test.ts.

Genie:  I started the wish workflow in the background on branch
        wish/delete-dead-defaults. It checks that the task fits a single wish,
        makes the change in its own worktree, runs the full check, gets an
        independent review and opens a PR against dev.

Genie:  Review: SHIP. Checks: pass. The PR is open and its remote head matches the
        local one. This run merged nothing. Merging is yours.

How it works

Designs and plans stay in your repository as markdown, with the reviews that approved them, so the next agent you use reads the same context. Each step is a skill. One decided task goes straight to wish. Anything bigger takes the whole loop, where every gate is run by an agent that did not do the work:

brainstorm → design review → wish → plan review → work → implementation review → PR

  • brainstorm settles an idea into a DESIGN.md. A design review approves it before any plan is written.
  • wish delivers one task end to end and returns a pull request ready for you to merge. Work bigger than one task becomes a WISH.md with execution groups, and a plan review approves it before anything is built.
  • work runs the approved groups in dependency order, each with its own scoped worker, and ends at an implementation review.
  • review runs those reviews. After work, it checks the result against the plan's criteria and returns SHIP, FIX-FIRST or BLOCKED. fix repairs the blocking gaps with bounded retries and a fresh reviewer.
  • council sits beside the loop. It pressure-tests a decision through independent lenses and changes nothing.

Designs, plans and verdicts land in git under .genie/brainstorms// and .genie/wishes//. Task state lives in one SQLite file per repository, .genie/genie.db, and genie board shows it as a kanban. Genie is built with Genie: this repository's own designs and plans are in .genie/brainstorms and .genie/wishes.

Skills

Genie v6.261003.1 ships 18 skills. The skill catalog groups them by category, and the Skills docs show a card for each one.

All 18 skills

  • authoring: Write a Genie skill that passes the shipped contract
  • brainstorm: Settle an idea into a reviewed design
  • council: Pressure-test decisions through independent lenses
  • deslop: Clean up scoped prose, code or interface without losing meaning
  • docs: Audit docs and DX against the live product; write on request
  • fix: Repair review gaps with bounded retries
  • genie: Route or resume work only where Genie adds value
  • genie-hacks: Find grounded Genie workflow patterns and recipes
  • merge: Resolve a merge or rebase by intent, then re-run the gate
  • refine: Improve prompts for OpenAI or Claude
  • report: Diagnose a failure to root cause; file issues on request
  • research: Investigate against primary sources and cite every claim
  • review: Assess plans, code, PRs, and repository quality
  • skill-audit: Audit the skill catalogue for overlap, staleness, drift
  • verify: Prove completion with fresh gate, diff, and check output
  • wish: Deliver one task end to end, or plan a multi-group wish
  • work: Execute approved wishes with evidence
  • workfly: Build a saved workflow from a procedure

Workflows

A saved workflow is a Claude Code script for a procedure worth running the same way twice. genie install and genie update deliver the catalog to ~/.claude/workflows. Genie v6.261003.1 ships 12. The Workflows docs describe each one. In Codex and other agents without saved workflows, the skills that start them (wish, brainstorm, council, docs, research, skill-audit, workfly) run the same stages by hand.

All 12 saved workflows

  • brainstorm: runs one round of a brainstorm whose state lives in DRAFT.md, so you answer between rounds, and ends in a reviewed DESIGN.md
  • council: architecture, delivery, product, security and dissent lenses, then a synthesis; assess only
  • docs-audit: audits every documentation surface against the live product and ranks the drift; assess only
  • evidence-gate: verifies a frozen contract of files, commands and claims against the artifact it covers; its only write is the report
  • observability-review: reviews traced Claude Code sessions in Phoenix and proposes rule changes, each tied to cited evidence; proposal only
  • pm-ledger-verify: adversarial lenses over uncommitted wish ledger edits
  • research-sweep: investigates a frozen question against a frozen source list and keeps every claim cited; read only
  • skill-audit-sweep: sweeps the shipped skills into a keep, improve, update, merge or retire table; assess only
  • skill-intake: assesses candidate skills from outside the catalog and assigns each exactly one disposition; mutates nothing
  • test-simplify: audits every test file in scope and proposes batches of tests to delete or consolidate; deletes nothing
  • wish: the engine behind the wish skill: admission, one executor in a worktree, a gate, an independent review, bounded repair and a PR; merging stays with you
  • workfly: discovers a procedure and builds its saved workflow, checked by a static test and refuters

When not to use it, and what it costs

Genie adds steps around the code. They pay off when a change deserves review, and they cost you time when it does not.

  • Small, obvious edits. The run above asked 0 questions and, by its own genie wish report, took 19.9 min and 387,079 tokens to delete dead code. That is one sample. An edit you can make and check by hand is faster without Genie.
  • Big plans. Plan size predicts delivery. In an observational sample of 50 wishes, the share merged within 8h fell from 100% (n=9) for the smallest plans to 95% (n=19), 67% (n=12) and 30% (n=10) for the largest. Split big work before it starts.
  • Diffs you will not read. Agents miss things, reviewers included. The reviewer passed a wrong gate command in #3045, and #2935 met 68 of 86 blind criteria with 3 HIGH gaps. Read the diff and the verdict before you merge.
  • Weak checks. The gate runs your repository's checks, so a wish proves only what those checks prove.
  • Untrusted repositories, or no integration branch. /wish runs the repository's own install and check commands, so point it only at repositories you trust. It opens its pull request against dev by default and refuses main or master as the base.

How it compares

Superpowers, Spec Kit and GSD work the same ground. Each of their rows comes from that project's own README.

What it says it is Its loop How it installs
Superpowers "a complete software development methodology for your coding agents, built on top of a set of composable skills" brainstorming, writing plans, then executing them with subagents or inline, with code review and test-driven development installed per harness, "separately for each one"
Spec Kit "an open source toolkit that gives AI coding agents structured processes, reusable templates, and documented outcomes" a constitution once per project, then specify, plan, tasks, implement and converge for each feature uv tool install specify-cli
GSD "a context-engineering and spec-driven development framework" that runs heavy work "in fresh-context subagents" discuss, plan, execute, verify and ship, one phase at a time npx @opengsd/gsd-core@latest
Genie skills and saved workflows that keep plans in git, plus a CLI that keeps task state in one SQLite file brainstorm, wish, work and review, with council beside it and a separate reviewer at each gate one signed binary that installs the skills into every agent it detects

Install

The one-paste install at the top fetches the signed binary for Linux or Apple Silicon macOS and finishes with genie install. Stable is declared for Linux and macOS: both run in the release gate, so a change that breaks either one does not ship. Windows is not supported; WSL2 works but is not on the tested matrix. Coming from an earlier version? UPGRADING.md covers what changed, what left, and how to clean up.

What the install delivers, consent scopes and release verification

Genie ships exactly two surfaces:
- The signed binary, installed and updated by install.sh and genie update.
- The skills and the saved workflows, delivered through the skills.sh channel. genie install and genie update run the pinned skills CLI over the tree the signed release put on disk, copy the workflows into ~/.claude/workflows, and record what landed in ~/.genie/skills-install.json.

The installer hands its arguments to genie install. --integrations auto · codex · claude · all · none (or --skip-integrations) is the consent scope for the skills channel: any value other than none installs to every detected agent skill home, because the skills CLI already installs per agent; none skips the channel, writes no record, and reports skills: skipped (consent: none). To install the binary and skip the skills channel:```bash
curl -fsSL https://raw.githubusercontent.com/automagik-dev/genie/main/install.sh | bash -s -- --integrations none


Each detected agent gets the skills in its own skills home: `~/.claude/skills` for Claude Code, and the shared `~/.agents/skills` for Codex and the other agents that read it. `genie install` and `genie update` install the skills from the tree the signed release delivered, never from a GitHub ref. The public `npx skills add automagik-dev/genie` command serves the repository's default branch instead, so it can be ahead of or behind any release.

Every release is cosign-signed with SLSA provenance, and `genie update` verifies it offline, with no GitHub credential. The repository-hosted `.well-known/latest.json` and `dev.json` manifests are the authoritative channel pointers. GitHub's `/releases/latest` route and prerelease badge are deliberately not channel authority: a promotion advances only a monotonic manifest and never rewrites already-published assets or channel-significant draft/prerelease/latest metadata. Security and releases has the details.

## Commands

Genie v6.261003.1 has 17 CLI commands. Nothing runs in the background: every command does its work and exits. The CLI reference has the full `--help` output.

All 17 commands

| Command | What it does |
|---------|-------------|
| `genie board` | Kanban view of task state, derived live by query |
| `genie config` | Read the resolved global config, for example `config get budgets.maxEscalationsPerGroup` |
| `genie context` | Resolve spawn context as versioned JSON: the wish or group branch and base SHA, or the integration branch |
| `genie doctor` | Check the installation; `--fix-global-db` repairs a contaminated machine-scope database, backup first |
| `genie help` | Show help for any command |
| `genie idea` | Capture an idea into the roadmap board's Idea lane |
| `genie init` | Set up per-repository state and remove Genie's own historical MCP entries, backup first |
| `genie install` | Finish a verified install and deliver the skills under the consent scope you chose |
| `genie metrics` | Opt-in lifecycle capture, off until `metrics enable`: a machine-local ledger of card events with the runtime session they ran in; `metrics export` reports time and tokens per lifecycle transition and, with `--phoenix`, sends them to a Phoenix you name; `metrics prices update` fetches an optional model price table so USD can be derived where the runtime did not price a call |
| `genie mikro` | Run and grow mikro microagents in any repository; `mikro call ` returns validated JSON with every citation checked |
| `genie orca` | Retired: a stub for one more release that prints a notice and writes nothing |
| `genie setup` | Configure Genie |
| `genie shortcuts` | Manage tmux keyboard shortcuts |
| `genie task` | Inspect and drive task state |
| `genie uninstall` | Remove the Genie CLI, its `~/.genie` home (backups kept), its client plugin registrations and the skills and workflows it recorded |
| `genie update` | Update to the latest signed release, verified offline |
| `genie wish` | `wish lint` checks any repository's wishes for structure; `wish report ` prints a run's tokens and time per stage |

## Repository packages

Mikro lives in `mikro/` with its original Git history, npm lockfile, Node build and independent CI. Its runtime and dependencies are not part of the Genie binary or release payload. Root Bun, Biome and Knip discovery exclude that package; its full npm checks run separately.

The standalone runtime requires Node 22.19 or newer and pins the Pi SDK family to 1.0.2.
`genie mikro call  --engine rlm|pi` selects an engine explicitly; otherwise the
agent's `engine` wins, then `rlm`. RLM remains the default. Pi uses an isolated native
session with read, grep, glob, hermetic git and `emit_done`, not the RLM Python tool pack
or an ambient Pi installation. Both paths retain the caller's schema and citation checks.
Missing or partial provider usage is unknown: an observed subtotal is not a complete bill.
See the agent schema and
standalone upgrade boundary.

Repository checks use Bun 1.3.11 or newer, matching CI: subtree test pruning relies on `test.pathIgnorePatterns`. This development prerequisite does not change Genie's existing CLI runtime floor.```bash
bun run check
npm --prefix mikro run deps:ci
npm --prefix mikro run build
npm --prefix mikro run check
npm --prefix mikro test

mikro update is for a standalone Mikro Git worktree, including linked worktrees. It refuses the nested package before acquiring update ownership or changing Git state. Update the owning Genie source checkout explicitly instead; genie update remains the signed installed-release update path.

Links

Docs: Skills, Workflows, CLI reference, Security and releases · UPGRADING.md · Releases · automagik-dev, the org behind Genie · Discord · MIT License

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
genie 怎么安装?

在客户端按下面的安装命令接入:npx -y genie。对应的包名是 genie。装完在客户端配置里引用同一个包或端点。

genie 怎么在 Claude Desktop、Cursor、Cline 里配置?

它的传输方式是stdio · 本地进程,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

genie 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

genie 需要 API Key 吗?

资料没有标注鉴权方式,是否需要密钥取决于服务本身,接入前按官方文档确认。

genie 支持哪些客户端?

资料标注支持 Claude Desktop、Cursor、Cline、Claude Code;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

genie 能提供哪些能力?

资料显示能力标签包括 genie、ai-agents、ai-developer-tools、anthropic、automation。具体工具清单以官方文档为准。

genie 需要什么运行环境?

运行时为 Cross-platform;属于本地进程,需要在本机装好运行环境。

genie 是本地还是远程?

它是本地进程(stdio),由客户端在本机拉起。传输方式为stdio · 本地进程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-04-10。