agentic-identity-lab

MCP运维云端

提供方 jlunde-pg · 包名 mcp-remote

运维云端 stdio · 本地进程 OAuth 授权 社区 / 第三方 Claude DesktopClaudeZed
查看接入方式

An AI agent permission management service based on OAuth 2.0 and RBAC to constrain AI agent's access to user files.

一句话选型结论

按资料字段整理

agentic-identity-lab 属于运维云端类 MCP 服务器,stdio · 本地进程,OAuth 授权,可直接接入 Claude Desktop、Claude、Zed,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。

分类运维云端 传输方式stdio · 本地进程 鉴权OAuth 授权 部署方式本地部署
GitHub 星标 1 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

运维云端
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

Local Deployment Permission Management AI Agent Python OAuth 2.0 RBAC File Access Control

支持客户端

适用场景

Local DeploymentPermission ManagementAI Agent

安装与配置

stdio · 本地进程 OAuth 授权
客户端在本机拉起子进程,数据不出本机
npx -y mcp-remote http://localhost:3000/sse

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 python,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
云资源查询

查实例、存储、账单等云上资源状态。

部署与发布

把部署、重启、回滚这类操作接进对话。

监控与告警

拉取监控指标,异常时让模型帮你判断。

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、Claude、Zed,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。
  • 使用前要准备好OAuth 授权对应的凭据,并按最小权限配置。
  • 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

  • 需要在运维云端场景里让模型直接调用外部能力的人。
  • 已经在用 Claude Desktop、Claude 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。
  • 不方便管理密钥或第三方授权的人。
  • 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
agentic-identity-lab当前 未标注 stdio · 本地进程 本地部署 有 2026-02-07
lobehub 未标注 stdio · 本地进程 本地部署 有 2026-04-11
TrendRadar 未标注 stdio · 本地进程 本地部署 有 2026-04-13
Ui Tars Desktop 8a825770 未标注 stdio · 本地进程 本地部署 有 2026-05-17
ruflo 未标注 stdio · 本地进程 本地部署 有 2026-04-13
FastGPT 未标注 stdio · 本地进程 本地部署 有 2026-04-11

服务器档案

运维云端
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jlunde-pg上游标注的维护方 jlunde-pg
分类运维云端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OAuth 授权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mcp-remote · python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2026-02-07上游最后核验时间

详细介绍

9.1 KB
来自项目仓库与上游目录的原文,未做删改

🔮 The Palantír

Non-Human Identity with Per-User Permission Scoping for AI Agents

A proof-of-concept MCP server demonstrating how to scope an AI agent's file access to a user's permission boundaries using OAuth 2.0 delegated authorization, Auth0 RBAC, and Google Drive.

> The Palantír sees only what its user is permitted to see.


The Problem

When an AI agent acts on behalf of a user, it needs a non-human identity (NHI) that is constrained to that user's permissions. Without proper scoping, a single agent becomes a skeleton key to all organizational data.

This lab demonstrates the correct pattern: the agent authenticates users through an identity provider, receives a token containing only that user's permissions, and uses those permissions to constrain which resources it can access.

Architecture```

┌─────────────┐ ┌───────────────────────┐ ┌──────────────┐
│ Claude │────▶│ Palantír MCP Server │────▶│ Auth0 │
│ Desktop │◀────│ (Python · FastMCP) │◀────│ (IdP) │
└─────────────┘ │ │ └──────────────┘
│ Session: │
│ • user identity │ ┌──────────────┐
│ • auth0 permissions │────▶│ Google Drive │
│ • google credentials │◀────│ (Files) │
└───────────────────────┘ └──────────────┘


**The key insight:** Authorization comes from Auth0 (what can this user access?), not from Google Drive. The agent's Drive token is broad, but the MCP server only queries folders the user's Auth0 permissions allow.

## Demo: Same Agent, Different Users, Different Access

### Frodo (Hobbit Role) → 5 files```
📂 Shire Files (read:shire-files)
  📄 shire-map.txt
  📄 ring-bearer-journal.txt
  📄 second-breakfast.txt
📂 Shared Fellowship (read:shared-files)
  📄 council-of-elrond.txt
  📄 fellowship-roster.txt
❌ Mordor Files → INVISIBLE

Gandalf (Wizard Role) → 8 files```

📂 Shire Files (read:shire-files)
📄 shire-map.txt
📄 ring-bearer-journal.txt
📄 second-breakfast.txt
📂 Mordor Files (read:mordor-files)
📄 mordor-intel.txt
📄 orc-deployments.txt
📄 mount-doom-route.txt
📂 Shared Fellowship (read:shared-files)
📄 council-of-elrond.txt
📄 fellowship-roster.txt


Same agent. Same Google Drive. **Auth0 JWT permissions determine what the agent can see.**

## How It Works

1. **User calls `login`** → MCP server opens browser for Auth0 authentication
2. **Auth0 returns a JWT** with a `permissions[]` claim based on the user's role
3. **User calls `connect_google_drive`** → OAuth consent for Drive read access
4. **User calls `search_files` or `list_my_files`** → MCP server maps Auth0 permissions to folder IDs and only queries authorized folders
5. **User calls `read_file`** → MCP server verifies the file's parent folder is in the user's authorized set before returning content

## Permission Model

| Auth0 Permission    | Google Drive Folder     | Description              |
|---------------------|-------------------------|--------------------------|
| `read:shire-files`  | Shire Files folder      | Hobbit homeland docs     |
| `read:mordor-files` | Mordor Files folder     | Enemy intelligence       |
| `read:shared-files` | Shared Fellowship folder| Fellowship-wide docs     |

| Role   | User    | Permissions                                              |
|--------|---------|----------------------------------------------------------|
| Hobbit | Frodo   | `read:shire-files`, `read:shared-files`                  |
| Wizard | Gandalf | `read:shire-files`, `read:mordor-files`, `read:shared-files` |

## Setup

### Prerequisites

- Python 3.11+
- Node.js (for `mcp-remote` bridge)
- Auth0 account (free tier)
- Google Cloud project with Drive API enabled
- Claude Desktop

### 1. Clone and install```bash
git clone https://github.com/YOUR_USERNAME/palantir-mcp-server.git
cd palantir-mcp-server
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Auth0 Setup

  1. Create an Auth0 tenant
  2. Create an API called "The Palantir" with identifier https://palantir.middle-earth.local
    - Enable RBAC
    - Enable "Add Permissions in the Access Token"
  3. Add permissions to the API: read:shire-files, read:mordor-files, read:shared-files
  4. Create roles:
    - Hobbit → read:shire-files, read:shared-files
    - Wizard → all three permissions
  5. Create users and assign roles (e.g., Frodo → Hobbit, Gandalf → Wizard)
  6. Create a Regular Web Application ("Palantír MCP Agent")
    - Note the Client ID and Client Secret
    - Set Allowed Callback URLs: http://localhost:3001/callback

3. Google Drive Setup

  1. Create a Google Cloud project and enable the Google Drive API
  2. Create OAuth 2.0 credentials (Web application)
    - Authorized redirect URI: http://localhost:3002/callback
  3. Create three folders in Google Drive with test files:
    - Shire Files/ — hobbit-related documents
    - Mordor Files/ — enemy intelligence
    - Shared Fellowship/ — fellowship-wide documents
  4. Note each folder's ID from the Drive URL

4. Configure environment```bash

cp .env.example .env


Edit `.env` with your credentials:```
AUTH0_DOMAIN=your-tenant.us.auth0.com
AUTH0_CLIENT_ID=your_client_id
AUTH0_CLIENT_SECRET=your_client_secret
AUTH0_AUDIENCE=https://palantir.middle-earth.local

GOOGLE_CLIENT_ID=your_google_client_id
GOOGLE_CLIENT_SECRET=your_google_client_secret

GDRIVE_SHIRE_FOLDER_ID=your_folder_id
GDRIVE_MORDOR_FOLDER_ID=your_folder_id
GDRIVE_SHARED_FOLDER_ID=your_folder_id

5. Run the server```bash

python server.py


### 6. Configure Claude Desktop

Add to your `claude_desktop_config.json`:```json
{
  "mcpServers": {
    "palantir": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "http://localhost:3000/sse"]
    }
  }
}

Restart Claude Desktop.

7. Test it

In Claude Desktop:
1. "Log in to the Palantír" → authenticate as Frodo or Gandalf
2. "Connect my Google Drive" → authorize Drive access
3. "List my files" → see only what your role permits
4. "Search for ring" → scoped search across authorized folders
5. Log out, switch users, and see different results

MCP Tools

Tool Description
login Authenticate via Auth0 (opens browser)
connect_google_drive Link Google Drive (opens browser for OAuth consent)
whoami Show current user, permissions, and session status
list_my_files List all files in authorized folders
search_files Search across authorized folders
read_file Read a specific file (with authorization check)
logout Clear session and tokens

Key Architectural Principles

  • Least privilege by default — the agent has zero standing permissions; all access derives from the user's token
  • Delegation, not impersonation — the identity chain is preserved (agent acting on behalf of user)
  • Authorization ≠ authentication — Auth0 handles what you can access; Google handles file storage access
  • Short-lived, scoped tokens — no long-lived API keys or shared secrets
  • Defense in depth — read_file verifies parent folder authorization even if a file ID is guessed

Anti-Patterns This Avoids

  • ❌ God-mode service accounts with app-layer filtering
  • ❌ Shared API keys across all users
  • ❌ Client-side only permission enforcement
  • ❌ Long-lived credentials or hardcoded secrets

Production Considerations

This is a lab/POC. For production, you would want:

  • Token refresh — handle expired Google and Auth0 tokens
  • Proper JWT verification — validate Auth0 token signatures using JWKS
  • Persistent session store — Redis or database instead of in-memory
  • Multi-user concurrency — session isolation per concurrent user
  • Dynamic permission mapping — lookup folder mappings from a database instead of static config
  • Audit logging — log every file access with user identity and timestamp

Tech Stack

Component Technology Cost
Identity Provider Auth0 (free tier) Free (7,500 users)
File Storage Google Drive API Free (15 GB)
Agent Runtime Python + MCP SDK Free (open source)
AI Client Claude Desktop Free (with Claude plan)

License

MIT

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
agentic-identity-lab 怎么安装?

在客户端按下面的安装命令接入:npx -y mcp-remote http://localhost:3000/sse。对应的包名是 mcp-remote。装完在客户端配置里引用同一个包或端点。

agentic-identity-lab 怎么在 Claude Desktop、Claude、Zed 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。

agentic-identity-lab 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。

agentic-identity-lab 需要 API Key 吗?

资料标注需要OAuth 授权认证,使用前准备好对应凭据,并按最小权限配置。

agentic-identity-lab 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Zed;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。

agentic-identity-lab 能提供哪些能力?

资料显示能力标签包括 Local Deployment、Permission Management、AI Agent、Python、OAuth 2.0、RBAC。具体工具清单以官方文档为准。

agentic-identity-lab 需要什么运行环境?

运行时为 python;属于本地进程,需要在本机装好运行环境。

agentic-identity-lab 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 2026-02-07。