ParentSquare MCP
MCP server that gives Claude read access to your ParentSquare parent account, enabling queries about school posts, calendars, messages, directory, signups, forms, payments, and files. It also parses image/PDF attachments so calendar flyers are accessible.
README
ParentSquare MCP
An MCP server that gives Claude read access to your own ParentSquare parent account — school posts, calendars, messages, the staff directory, signups, forms, payments and files.
School calendars and flyers are very often posted as image or PDF attachments rather than as calendar entries. This server downloads them and hands them to Claude as real image/text content, so "what's on the school calendar next week?" works even when the calendar itself is empty.
Before you start: how this connects
ParentSquare has no public parent-facing API. Their documented API is district-side, for pushing student data in from an SIS — it is issued to district IT, not to parents, and it is not what you want here. So this server signs in as you and reads the same parent web app you use in a browser.
Two consequences worth knowing up front:
- You need to supply your own session. See Authentication.
- The page layouts are not a contract. Districts run different versions of the app on different
URLs, so there are three layers of defence: every route is a list of candidate paths tried in
order; if all of them fail, the section is located by matching its label in your sidebar; and
if a parser still finds nothing, it returns the readable page text rather than failing.
scripts/doctor.pyreports what needed which. See Fitting it to your district.
Install
cd "~/Desktop/ParentSquare MCP"
python3 -m venv .venv
./.venv/bin/python -m pip install -e .
Authentication
Method 1 — browser cookie (recommended)
Works with every district, including ones that sign in through Google, Clever or ClassLink, and with two-factor accounts.
- Sign in to ParentSquare in your browser.
- Open DevTools → Network, click any request to
parentsquare.com. - Under Request Headers, copy the entire value of
Cookie:. - Put it in
PARENTSQUARE_COOKIE(see Connect it to Claude).
The cookie expires eventually — when tools start reporting that the session expired, repeat this.
Method 2 — email and password
Only works if your district uses a plain ParentSquare password, with no two-factor.
PARENTSQUARE_EMAIL=you@example.com ./.venv/bin/python scripts/login.py
This caches the session in ~/.parentsquare-mcp/session.json (mode 600). If your district uses SSO
or 2FA, the script says so and points you back to Method 1.
Connect it to Claude
Claude Desktop
Edit ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"parentsquare": {
"command": "/Users/chen/Desktop/ParentSquare MCP/.venv/bin/python",
"args": ["-m", "parentsquare_mcp"],
"cwd": "/Users/chen/Desktop/ParentSquare MCP",
"env": {
"PARENTSQUARE_COOKIE": "_parentsquare_session=…; remember_user_token=…"
}
}
}
}
Restart Claude Desktop.
Claude Code
claude mcp add parentsquare \
--env PARENTSQUARE_COOKIE="_parentsquare_session=…" \
-- "/Users/chen/Desktop/ParentSquare MCP/.venv/bin/python" -m parentsquare_mcp
If you used scripts/login.py, the cached session is picked up automatically and you can drop the
env block entirely.
See .env.example for every setting.
Fitting it to your district
Run this once after connecting:
./.venv/bin/python scripts/doctor.py
It reports, for every section, which URL works and how many rows the parser got:
— routes —
ok feed /feeds
ok directory /directory
NAV payments /school_pay (found via the sidebar)
MISS volunteer_hours (no candidate worked, and no sidebar link matched)
— tools (23 registered) —
ok get_feeds 12 rows
THIN list_signups 0 rows — selectors need work. Source: https://…/signups
- NAV — none of the guessed URLs worked, so the server found the section by matching its
sidebar label instead. This already works; adding the reported path to
routes.pyjust saves a lookup. Districts re-route these pages constantly, which is why label matching is the safety net. - MISS — no URL worked and no sidebar link matched. Open the section in your browser, copy the
path from the address bar, and add it to the front of the matching list in
parentsquare_mcp/routes.py. If the sidebar calls it something unusual, adding that word toSECTION_KEYWORDSin the same file fixes it for good. - THIN — the page loaded but nothing matched. Ask Claude to call
debug_fetchon that URL to see the real markup, then adjust the selector list at the top of the relevant file inparentsquare_mcp/tools/. The selectors are plain CSS lists, deliberately easy to edit. - empty — the page loaded and genuinely has nothing in it. Nothing to fix.
Nothing in the doctor writes to ParentSquare.
Tools
Feed and posts
| Tool | What it does |
|---|---|
get_feeds |
Paginated feed: titles, authors, dates, summaries, attachment names. Takes a query filter. |
get_post |
Full post — body, comments, poll results, signup items, and the contents of attached images and PDFs. |
get_group_feed |
Posts from one group. |
Calendar
| Tool | What it does |
|---|---|
get_calendar_events |
Structured events from the ICS feed, recurring events expanded. When empty, explains how to find calendars posted as attachments. |
Communication
| Tool | What it does |
|---|---|
list_conversations / get_conversation |
Read message threads. |
get_directory |
Staff directory: name, role, phone, email, user id. |
get_staff_member |
Full profile including office hours and profile photo. |
Media and files
| Tool | What it does |
|---|---|
list_photos |
Photo gallery with image URLs. |
list_files |
Shared documents with download URLs. |
download_file |
Save any attachment to disk. |
Things to act on
| Tool | What it does |
|---|---|
list_signups |
Sign-ups and RSVPs with progress (e.g. 53/103 Items). |
list_notices |
Alerts and secure documents. |
list_polls |
Polls with vote counts and the leading option. |
list_forms |
Permission slips, with an outstanding count. |
list_payments |
Payment items with prices and totals. |
list_volunteer_hours |
Logged hours with the total. |
Discovery
| Tool | What it does |
|---|---|
list_schools |
Schools and students, with the ids other tools take. |
list_school_features |
Which sections a school actually has, from its sidebar. |
list_groups |
Groups with member counts and membership status. |
list_links |
Pinned quick links. |
get_student_dashboard |
A student's school, grade, classes and teachers. |
debug_fetch |
Escape hatch: fetch any ParentSquare path and see what is really there. |
Nothing here writes to ParentSquare — every tool is read-only against your account. On your own disk,
download_file saves where you ask it to, and get_post saves image-only PDFs (scans, which have no
extractable text) to ~/Downloads/ParentSquare so they are still reachable.
Things to try
What did the school send home this week?
Is there anything I still need to sign or pay for?
The October calendar was posted as a PDF — what days is my kid off?
Who is my son's teacher and what's their email?
Tests
./.venv/bin/python tests/run_all.py
Four suites — parsers, tools over HTTP, a real MCP handshake over stdio, and the authentication paths — all against a fake ParentSquare served on localhost. They never touch the real site, so they are safe to run at any time.
Layout
parentsquare_mcp/
config.py environment settings
client.py auth, HTTP, route fallback, login-bounce detection
routes.py candidate URLs + sidebar keywords ← edit when doctor says MISS
parsers.py shared HTML helpers and the fallback result shape
media.py attachments → images / PDF text
tools/ one module per group of tools ← edit when doctor says THIN
scripts/
login.py cache a session with email + password
doctor.py check routes and tools against your account
tests/
Notes
- Read-only by design. Nothing posts, replies, signs, pays or RSVPs on your behalf.
- Your credentials stay on your machine: in the client config env block, or in
~/.parentsquare-mcp/session.jsonat mode 600. - Because it depends on the parent web app's HTML, a ParentSquare redesign can break parsing. That is
what
doctor.pyanddebug_fetchare for, and why nothing hard-fails when a selector misses.
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。