aware-connector
MCP server for Autodesk's Aware employee directory, enabling search for people by name, email, or title, fetching individual records, viewing org reporting lines, and reading your own profile. It uses your own logged-in session to access the same data as the web app, read-only.
README
aware-connector
Read Autodesk's Aware employee directory from your terminal, or expose it to an LLM as an MCP server.
Aware has no public API and no official MCP integration. This connector talks to the same endpoints the web app does, using your own session, and returns only data you are already authorized to see.
$ aware search "principal engineer" --limit 2
Jane Doe · Principal Engineer, Platform Services
jane.doe@autodesk.com · Portland, US
manager: John Roe · org: Platform Services Engineering
Sam Vale · Principal Engineer, Fusion
sam.vale@autodesk.com · Toronto, CA
manager: Alex Kim · org: Fusion Core
Showing 2 of 47 matches. Refine the query, or use --limit N / --all.
Requirements
- macOS. The refresh token is stored with the
securitykeychain CLI, andpbpasteis used for the paste-based login. - Node 24 or newer — the active LTS, and the floor CI tests against.
Development happens on Node 26, which
mise.tomlpins. TypeScript runs directly via Node's native type-stripping, so there is no build step and nothing is compiled.
Install
git clone https://github.com/camilopiedra92/aware-connector.git
cd aware-connector
npm install # only needed for the MCP server; the CLI has no runtime deps
npm link # optional, puts `aware` on your PATH
Without npm link, run commands as node src/cli.ts <command>.
If you manage Node with a version manager (mise, nvm, fnm), be aware that
npm link writes into the npm prefix of the exact version that is active, so
the command disappears the next time you upgrade Node. A small wrapper on your
PATH that calls node /path/to/src/cli.ts "$@" survives that, and lets you pin
which runtime the CLI gets rather than inheriting the current directory's.
Logging in
Aware issues tokens through an AWS Cognito user pool federated to Azure AD. The connector stores a long-lived refresh token and mints short-lived access tokens on demand, so you do this once.
- Open https://one.autodesk.com/apps/aware/ in Chrome, signed in.
- DevTools (
Cmd+Opt+I) → Application → Cookies →https://one.autodesk.com. - Copy the value of the cookie whose name ends in
.refreshToken(CognitoIdentityServiceProvider.<clientId>.<user>.refreshToken). - Paste it into
aware login:
aware login # prompts, echo off
pbpaste | aware login # or straight from the clipboard
The token is validated against Cognito before being written to your login
keychain. It never touches the repository, a dotfile, or an environment
variable. aware logout removes it.
Bootstrap is paste-only on purpose. An earlier version ran a local HTTP endpoint for the browser to POST the token to; it was removed because an unauthenticated localhost endpoint that stores a credential is CSRF-vulnerable — any open web page could plant a token in your keychain.
CLI
| Command | What it does |
|---|---|
aware login |
Store your Cognito refresh token in the keychain |
aware me |
Your own directory record |
aware search <query> |
Match people by name, email, or job title |
aware people |
The full org feed (~20k people) as JSON on stdout |
aware logout |
Remove the stored credential |
Flags: --limit N (default 20) and --all for search; --refresh on any
read to bypass the local cache and re-fetch.
search writes records to stdout and counts to stderr, so aware people > org.json
gives you clean JSON with the summary still visible in the terminal.
MCP server
src/mcp.ts speaks JSON-RPC over stdio and exposes four read-only tools:
| Tool | Purpose |
|---|---|
aware_search |
Search by name, email, or title; reports the true match count when results are capped |
aware_get_person |
One person, by exact work email |
aware_me |
The signed-in user's own record |
aware_org |
The reporting line around a person: managers up, direct reports down |
Register it with Claude Code:
claude mcp add aware -- node /absolute/path/to/aware-connector/src/mcp.ts
Or, for any client that reads a JSON config:
{
"mcpServers": {
"aware": {
"command": "node",
"args": ["/absolute/path/to/aware-connector/src/mcp.ts"]
}
}
}
All four tools are annotated readOnlyHint — nothing here writes to Aware.
Run aware login first; the server reads the same keychain entry the CLI does.
How it works
Two reverse-engineered facts drive the whole design:
- Auth is AWS Cognito, not Autodesk Platform Services. Access tokens last an hour and are minted from the stored refresh token as needed.
- People data is a static CDN feed, not a REST API — a Workday export on
CloudFront, read with a Bearer token. There is no backend search, so
searchfilters the feed client-side exactly as the web app does. The/api/v1endpoints exist but use AWS IAM (SigV4) auth and carry no org data, so they are deliberately not implemented.
The feed is ~27 MB, so it is cached under ~/Library/Caches/aware-connector
with a 30-minute TTL, matching Aware's own Workday refresh cadence. Cache
failures are non-fatal; --refresh skips it.
Development
npm test # node --test over test/
node --test test/auth.test.ts # one file
node --test --test-name-pattern="refresh" # by test name
npm run typecheck # tsc --noEmit; the only "build"
npm run lint # eslint . — type-aware rules only
There is no bundler: nothing is shipped over a network, so the CLI runs from source and stack traces need no source maps. Tests cover the parts with real edge cases — token refresh timing, 401-retry, search and email filtering, org-chart cycles, cache-vs-network — plus a suite that drives the real MCP server over an in-memory transport, so a renamed tool or a malformed schema fails the build.
CI runs all of that on macOS against Node 24 and 26, and separately runs the CLI
on a bare checkout with no npm install, which is what keeps the "the CLI has no
runtime dependencies" claim honest.
Disclaimer
Unofficial and unaffiliated with Autodesk. It reads, with your own credentials, the same directory the Aware web app already shows you; it holds no special access and writes nothing. Whether running it fits your employer's acceptable-use policy is on you.
License
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。