maimai-cli

Unofficial CLI helpers and an optional Codex/agent skill for user-authorized Maimai web sessions.

maimai-cli is designed for interactive, low-volume reading of Maimai content that the signed-in user can already view in the browser. The repository contains two pieces:

• maimai, a local command-line tool published on PyPI.
• skill/, a routing skill that teaches an agent when and how to use the CLI safely.

It does not provide login bypasses, QR login, browser-cookie extraction, CAPTCHA solving, anti-bot bypasses, or bulk scraping workflows.

中文简介

maimai-cli 是一个非官方的脉脉本机命令行工具，用来读取当前登录用户本来就能在网页端看到的内容，例如推荐流、热榜、同事圈、搜索结果、帖子详情、评论、图片和资料卡。

这个仓库同时提供：

• maimai CLI：发布在 PyPI 上的命令行工具。
• skill/：给 Codex/Agent 使用的路由技能，帮助 Agent 在用户提到脉脉、同事圈、推荐流、评论、Cookie 等场景时正确调用 CLI。

安全边界：

• 不支持扫码登录。
• 不自动读取浏览器 Cookie。
• 不绕过登录、验证码、风控或权限控制。
• 只适合低频、交互式地查看用户自己有权限访问的内容。
• 不建议把完整 Cookie 发到聊天里，推荐在本机终端通过环境变量或标准输入导入。

快速开始：

pip install maimai-cli
maimai --help
maimai import-cookie-header
maimai status
maimai feed --limit 10
maimai company-feed --limit 10

安装 Agent skill：

cp -R skill ~/.agents/skills/maimai-cli

Safety Model

• You must provide your own valid Cookie header from a browser session you control.
• The CLI does not read browser cookie stores automatically.
• Cookies are stored locally under ~/.maimai-cli/cookies.json with best-effort 0600 permissions.
• The CLI only uses same-origin Maimai web endpoints and user-visible pages.
• Do not use this project to access accounts, companies, feeds, posts, comments, or profiles that you are not authorized to view.
• Respect Maimai's terms, privacy expectations, rate limits, and applicable law.
• Keep request volume low. This tool is designed for interactive use, not continuous crawling.
• If a Cookie header is exposed in chat, rotate it by logging out or refreshing the browser session.

Install

From PyPI:

pip install maimai-cli
maimai --help

With uvx:

uvx maimai-cli --help

From source:

git clone https://github.com/mariotong/maimai-cli.git
cd maimai-cli
uv run maimai --help

Agent Skill

The repository includes a Codex/agent skill at skill/. Install it into your local skill root if you want agents to route Maimai-related requests to this CLI:

cp -R skill ~/.agents/skills/maimai-cli

The skill is intended to trigger when a user mentions Maimai, maimai, 职言, 同事圈, 推荐流, 热榜, 帖子详情, 评论, Cookie, or related troubleshooting.

Skill references:

User intent	Reference
Authentication, Cookie import, safety boundaries	skill/references/auth.md
Recommended feed, hot rank, company circle, pagination	skill/references/feeds.md
Search content and contacts	skill/references/search.md
Details, comments, images, profiles, short indexes	skill/references/detail.md
Troubleshooting and common workflows	skill/references/troubleshooting.md

Agent usage principles from the skill:

1. Verify maimai --help and maimai status before deeper workflows.
2. Start list reads with small limits, for example maimai feed --limit 5.
3. Treat 1, 2, and similar numeric values as short indexes only after checking recent list context.
4. Summarize output by default; use --json, --yaml, or --raw only when needed.
5. Never ask users to paste complete Cookie headers into chat.

Authentication

Import a Cookie header explicitly:

maimai import-cookie-header --cookie '<paste your Cookie header here>'

You can also use an environment variable or stdin:

MAIMAI_COOKIE='<paste your Cookie header here>' maimai import-cookie-header

Check local authentication evidence:

maimai status
maimai me
maimai cookies

Remove saved cookies:

maimai logout

Common Commands

Read a visible community feed:

maimai feed --type recommended --limit 20
maimai feed --type gossip --offset 20 --limit 20
maimai feed --page 2 --page-size 10

Read hot rank:

maimai hot-rank --limit 20

Read the current account's visible company circle feed:

maimai company-feed --limit 20

Read a specific company circle if you already know its webcid:

maimai company-feed <webcid> --limit 20

If automatic company-circle discovery fails with an HTTP 406/404 or an empty result, open the company circle in the browser and copy the webcid from the URL:

https://maimai.cn/company/gossip_discuss?webcid=<webcid>

Then pass it explicitly:

maimai company-feed <webcid> --limit 20

Search visible content and contacts:

maimai search "keyword" --limit 5
maimai search "keyword" --section gossips
maimai search "keyword" --section contacts

After any listing command, entries are saved as short indexes. Use those indexes for details, comments, images, and profiles:

maimai refs
maimai detail 1 --kind gossip
maimai comments 1 --kind gossip --limit 20
maimai images 1 --kind gossip
maimai profile 1

Use structured output for scripts:

maimai feed --type gossip --limit 5 --json
maimai hot-rank --yaml

Use --raw only when you explicitly need the full parsed response:

maimai company-feed --raw

Short Indexes

List commands save a local reference table at ~/.maimai-cli/refs.json. Follow-up commands can use the latest list's 1-based indexes instead of copying raw IDs:

maimai feed --type gossip --limit 10
maimai refs
maimai detail 1 --kind gossip
maimai comments 1 --kind gossip

If a short index points to the wrong item, run maimai refs to inspect the current cache. A later list command replaces the short-index context.

Comments use a separate cache at ~/.maimai-cli/comment_refs.json, so reading comments no longer replaces the post indexes from feed, hot-rank, company-feed, or search. Inspect comment indexes explicitly when needed:

maimai comments 1 --kind gossip
maimai refs --scope comments
maimai comments 1 --kind gossip --cid 1

Short indexes are not raw post IDs. They only refer to the latest local list cache. Running another post list command can replace that cache. If you use a raw gossip ID directly, pass its --egid; if you use a raw feed ID directly, pass its --efid:

maimai detail 36744330 --kind gossip --egid <egid>
maimai comments 36744330 --kind gossip --egid <egid>

Dynamic Action IDs

Some Maimai web features use Next.js server action IDs. The CLI ships with known defaults and keeps a local runtime cache at ~/.maimai-cli/actions.json. If an action fails, the client can scan currently visible pages for candidate action IDs, retry conservatively, and cache a working value.

This mechanism is for compatibility with normal web-page changes. It is not intended to bypass access controls or anti-abuse systems.

Troubleshooting

Check auth first:

maimai status
maimai cookies
maimai me

If auth looks stale:

maimai logout
MAIMAI_COOKIE='<paste your new Cookie header here>' maimai import-cookie-header
maimai status

If list follow-up commands fail:

maimai refs

If search pagination appears unchanged, note that the current Maimai web search endpoint may ignore offset/page parameters.

If maimai company-feed fails during automatic company-circle discovery, pass the browser URL's webcid explicitly:

maimai company-feed <webcid> --limit 20

Development

Run local checks:

uv run python -m py_compile maimai_cli/*.py maimai_cli/commands/*.py
uv run python -m unittest

Build the package:

uv build

Publish to PyPI:

UV_PUBLISH_TOKEN='<your PyPI API token>' uv publish

Project Status

Alpha. Endpoints and page structures may change without notice. Expect breakage and treat output as best-effort.

Disclaimer

This project is unofficial and is not affiliated with, endorsed by, or sponsored by Maimai. Use it only with accounts and data you are authorized to access.