@bendyline/gezel-mcp

中文文档
1.2.13发布于 2026/11/05MITTypeScriptES Module构建来源已验证近 30 天下载 2041

Gezel's Model Context Protocol server: gives gezels and other MCP clients access to memory, workspace files, artifacts, tasks, and team tools.

原始使用文档
文档目录5 节
  1. For other AI apps
  2. The full surface, with your own access
  3. Tool categories
  4. Entry points
  5. Stability

@bendyline/gezel-mcp

The Model Context Protocol server that gives gezels, the AI companions in Gezel, their hands.

Every tool a gezel can call lives here: memory search and save, workspace file operations, project artifacts, the shared document library, script and Playwright execution, task management, team and project management, image rendering, and history search.

Gezel runs this server for every gezel's chat. It also gives other AI apps (Claude Code, Codex, Cursor, Claude Desktop, VS Code) a small set of Gezel's tools; see For other AI apps.

npm install @bendyline/gezel-mcp

For other AI apps

gezel-mcp --harness <client> (claude-code, claude-desktop, codex, cursor or vscode) gives an app that already has its own file, shell and web tools what only Gezel has: search over the folder it works in (including the text of its Office documents), the person's shared library, memories and knowledge catalogs, Office documents read as markdown, image description and generation, and memory. Nine tools in all, with absolute paths throughout.

Set it up from Gezel (Settings → Integrations, or gezel mcp install <app>), which writes the app's entry for you. By hand, run the server under Node 24 or later from the folder you work in:

{
  "mcpServers": {
    "gezel": {
      "command": "npx",
      "args": ["-y", "@bendyline/gezel-mcp", "--harness", "claude-code"]
    }
  }
}

The server binds the Gezel project for the folder the app reports (MCP roots), or --folder <path>, or its working directory; --project <id> names one outright. It reads the owner credential from ~/.gezel/runtime/ once and trades it for a token confined to that project and to the choices the person made in Gezel, so the app's model never acts with the person's full access. If Gezel is not running, every tool says so instead of the server exiting.

The full surface, with your own access

Without --harness the server offers every tool a gezel has, acting with your own access, as the gezel command does. It speaks stdio, so any MCP client can run it. With Gezel running on the same computer:

{
  "mcpServers": {
    "gezel": {
      "command": "npx",
      "args": ["-y", "@bendyline/gezel-mcp"],
      "env": {
        "GEZEL_AGENT_ID": "<a gezel id from `gezel agent list`>",
        "GEZEL_PROJECT_ID": "default"
      }
    }
  }
}

There is no address, password, or certificate to configure. The server is a thin front end that calls back into the running Gezel service, and finds it on its own: Gezel picks its port, its sign-in token, and its TLS certificate at each launch and writes them to ~/.gezel/runtime/ (port, auth-token, cert.pem), readable only by you. The server reads them when it starts and follows Gezel across a restart, so a long-lived MCP client keeps working after Gezel updates. If Gezel is not running when the client starts the server, it exits with a message saying to open the Gezel app or run gezel start; start it, then reconnect the server from your MCP client.

Variable
GEZEL_AGENT_IDThe gezel the tools work as: whose memories they search and save, whose tasks they see.
GEZEL_PROJECT_IDThe project tools use when a call names none. Default default.
GEZEL_HOMEA Gezel home other than ~/.gezel.

Run this way, the server acts with your own access, as the gezel command does; GEZEL_PROJECT_ID picks a default, not a boundary.

gezel agent list (from @bendyline/gezel-cli) lists your gezels and their ids. The agent in that command and in GEZEL_AGENT_ID is a historical name: both refer to your gezels.

When Gezel starts the server itself, for a gezel's chat, it passes the connection directly: GEZEL_BASE_URL, GEZEL_TOKEN (a token it mints for that one session, confined to its project unless the gezel coordinates across projects), and GEZEL_CERT_PATH (the runtime/cert.pem that token's service is using). Setting GEZEL_BASE_URL yourself turns the lookup off, and keeping all three current across restarts is then up to you.

Tool categories

Memory · workspace files · project artifacts · shared documents · script and Playwright execution · package installation · team and project management · tasks · user questions · history search · image rendering.

Run the server and call tools/list for the authoritative, current inventory.

Entry points

SubpathContents
@bendyline/gezel-mcpServer construction helpers
@bendyline/gezel-mcp/lint-contractsThe exact input and output schemas the built-in tools publish, for checking tool contracts
@bendyline/gezel-mcp/dist/server.jsThe stdio server itself, also installed as the gezel-mcp command. The Gezel service launches this file for every chat, so it stays a supported path

Stability

Public API under semver. Tool names and argument schemas are part of that contract: removing a tool or narrowing an argument is a breaking change.

MIT © Bendyline

版本与发布记录

@bendyline/gezel-mcp 共发布 26 个版本,首次发布于 2026/08/11;过去 12 个月发布了 26 个版本。最新版本附带 npm provenance 构建来源证明,可追溯到公开的 CI 构建记录。

版本发布日期距今
1.2.13latest2026/11/053 天前
1.2.122026/11/044 天前
1.2.112026/10/2811 天前
1.2.102026/10/2712 天前
1.2.92026/10/2712 天前
1.2.82026/10/2415 天前

关于 @bendyline/gezel-mcp 的常见问题

@bendyline/gezel-mcp 是什么?

Gezel's Model Context Protocol server: gives gezels and other MCP clients access to memory, workspace files, artifacts, tasks, and team tools.

如何安装 @bendyline/gezel-mcp?

在项目目录运行 npm install @bendyline/gezel-mcp,或使用 pnpm add / yarn add / bun add。它还提供命令行工具 gezel-mcp,可以用 npx gezel-mcp 直接运行。

@bendyline/gezel-mcp 支持 TypeScript 吗?

支持。包内自带 TypeScript 类型声明(package.json 声明了 types 入口),安装后即可获得类型提示,无需再安装 @types 包。

@bendyline/gezel-mcp 支持 ESM 还是 CommonJS?

以 ES Module 形式发布,推荐使用 import 引入。在 CommonJS 项目中,Node.js 20.19+ / 22.12+ 可以直接 require() 不含顶层 await 的 ES Module,更早的版本需要改用动态 import()。

@bendyline/gezel-mcp 需要什么版本的 Node.js?

package.json 的 engines 字段要求 Node.js >=24.0.0。版本不符时 npm 默认只给出警告,开启 engine-strict 后会拒绝安装。

@bendyline/gezel-mcp 可以商用吗?

@bendyline/gezel-mcp 使用 MIT 许可。宽松许可:允许商业使用、修改和再分发,通常只需在分发时保留原版权与许可声明。以上为概括,不构成法律意见,请以许可原文为准。

@bendyline/gezel-mcp 的下载量有多少?

根据 npm 官方下载统计,@bendyline/gezel-mcp 在 2026/10/08 – 2026/11/06 这 30 天内被下载 2,041 次,日均约 68 次。下载量会把 CI 构建、镜像同步和作为其他包依赖的安装一并计入,适合横向比较热度,不能直接等同于使用人数。

@bendyline/gezel-mcp 还在维护吗?

最新版本 1.2.13 发布于 2026/11/05(3 天前);至今共发布 26 个版本,其中过去 12 个月发布了 26 个;首次发布于 2026/08/11。发布频率只是参考信号:功能稳定的工具可能很少需要更新,建议结合源码仓库的 issue 和提交记录判断。