Connect Your Tutoring Business to Claude, ChatGPT or Codex — Ask Your Own Data Anything

How do I connect my tutoring business to Claude, ChatGPT or Codex?

ClassQuill runs a hosted MCP server at mcp.classquill.com that any of these AI assistants can connect to directly. Add it as a custom connector in claude.ai or ChatGPT (you just log in), or point Claude Code, Codex or Composio at it with an API key from Settings → Developers. Once it’s connected, you can ask about your sessions, tutors, students, payments and invoices in plain English — the assistant reads the real answer from your ClassQuill data. It’s read-only: nothing you ask it can change your account.

What actually happens when you connect ClassQuill to an AI assistant?

mcp.classquill.com runs on the Model Context Protocol{target=“_blank”} — the open standard behind how AI assistants connect to outside tools. It isn’t a separate product with its own data. It’s a thin proxy in front of the same /v1 API ClassQuill’s own app runs on: the same access control, the same org-scoping, the same rate limits. Whatever your account is allowed to see through the API, the connected assistant can see, and nothing more.

The important part for anyone nervous about handing an AI assistant access to real business data: it’s read-only, and that’s not a policy promise, it’s how the tool list is built. Every tool the server exposes only performs a GET request; there are no write tools live. Nothing you ask it can create a session, send an invoice, or change a record.

It answers questions. It doesn’t act on them.

The two ways to connect, and which one is yours

Create your API key once (org admins only — Settings → Developers in ClassQuill) and reuse it across every tool in the second row. The key carries the same permissions as your login — no separate approval step per tool.

Connect claude.ai

1. In claude.ai, go to Settings → Connectors → Add custom connector.
2. Enter https://mcp.classquill.com/mcp and add it.
3. Claude opens a ClassQuill login. Sign in and approve access — the connector’s tools then appear in your tool list.

Custom connectors are available on claude.ai’s paid plans.

Connect ChatGPT

Enable Developer Mode in ChatGPT (needs a paid plan), add https://mcp.classquill.com/mcp as a custom connector, and log in the same way. ClassQuill isn’t listed in ChatGPT’s public Plugin Directory (or Claude’s own connector directory, for that matter) — this is a self-add custom connector on both platforms, not something you search for and install. That’s not a gap: the custom-connector path is the same one Composio, Zapier and most other MCP servers use, and it works identically either way.

Connect Claude Code

claude mcp add classquill --transport http https://mcp.classquill.com/mcp --header "Authorization: Bearer ei_live_xxxxxxxxx"

Connect Codex

[mcp_servers.classquill]
url = "https://mcp.classquill.com/mcp"

[mcp_servers.classquill.headers]
Authorization = "Bearer ei_live_xxxxxxxxx"

Connect Composio

{
  "name": "classquill",
  "url": "https://mcp.classquill.com/mcp",
  "headers": { "Authorization": "Bearer ei_live_xxxxxxxxx" }
}

Any other MCP client

Send your key as a Bearer token on each request to https://mcp.classquill.com/mcp. Cursor, Windsurf, and most other MCP-aware tools follow the same shape as the Codex/Composio examples above.

Prefer not to depend on the hosted endpoint?

There’s also a published npm package, equateit-mcp, that runs the identical tool set yourself — stdio by default, or --http if you want to host it locally. Same data, same read-only access, you just run the process.

What can you actually ask once it’s connected?

Everything the server exposes maps to a real part of your ClassQuill data — this is what it looks like to ask ChatGPT or Claude about your own tutoring business instead of opening a dashboard:

• Sessions — what’s scheduled, what happened, coverage gaps
• Tutors — who’s available, and per-tutor earnings
• Payments & invoices — what’s been paid, what’s outstanding
• Students & parents — student records, and a parent’s balance
• Curriculum — subjects, classrooms, student groups, lesson plans, homework, results
• Operations — mileage, expense claims, and an owner’s reports summary

The real starter prompts ClassQuill’s own team tested this against:

• “What tutoring sessions do I have this week?”
• “Which invoices are overdue?”
• “Show me this month’s earnings for my tutors.”
• “Find the lesson plan for [student].”
• “Where are my coverage gaps this week?”

Each of those is a single question that would otherwise mean opening the app, finding the right screen, and filtering a report by hand. Connected, it’s a sentence.

The one I actually ask most: “Which of my tutors is the best fit for this new lead?” No dashboard view joins a fresh lead’s details against your tutor list — you’d normally open the lead, then open tutors, then match subject and availability by hand. Ask it connected, and the assistant reads both the lead and the tutor roster in one pass and reasons across them for you: subject match, availability, who’s actually the right fit, not just who’s free.

Is this safe to connect?

Reasonable question to ask before you connect anything to your business data. The short answer: yes, and specifically because of what it can’t do.

• Read-only. No write tools are exposed — it cannot create, edit or delete anything in your account.
• Scoped to your organisation. Whoever’s logged in or holding the API key only sees that organisation’s data, same as the app.
• Same access control as everywhere else. Rate limits, permissions and audit trail are all enforced by the same API your own app runs on — this isn’t a separate, looser door into your data.
• Revocable any time. Disconnect the custom connector in claude.ai/ChatGPT settings, or delete the API key in Settings → Developers, and access stops immediately.

The way I’d actually answer it: the risk with connecting an AI assistant to your business data isn’t “can it see something it shouldn’t” — the org-scoping and access control already handle that, same as the app itself. The real risk with most AI integrations is write access — an assistant that can act on your data, not just answer questions about it, is the one that can genuinely go wrong (delete the wrong record, send an invoice it shouldn’t have). That’s exactly the risk this doesn’t carry.

It can’t create, edit or delete anything, so the worst case if it ever gets something wrong is a wrong answer — annoying, not damaging. That’s a much easier thing to trust on day one than “give an AI the keys,” and it’s why a Claude-for-small-business setup that only reads is a fundamentally safer starting point than one that can also act.

What this doesn’t do yet

It won’t create a session, send an invoice, or mark anything for you — every tool is read-only today. If you want to build something that acts on your ClassQuill data (not just answers questions about it), the full public API supports writes — the MCP connection is the read-only, ask-anything layer on top of it, not a replacement for it. For fully automatic, rule-based actions (log every completed session to a spreadsheet, that kind of thing), that’s what ClassQuill’s no-code automation via webhooks and Zapier is for (link on publish) — different tool, different job: MCP is for when you’re asking, automation is for when nothing should require asking at all.

That’s a deliberate line, not a gap we haven’t gotten to yet. Handing an AI assistant the power to act on real student, payment and family data is a different risk calculus to handing it the power to answer questions about it — and it’s one worth being genuinely careful about before shipping, not one to rush past for the sake of a longer feature list.

Getting started

If you haven’t already, create an API key in Settings → Developers, or just add https://mcp.classquill.com/mcp as a custom connector in claude.ai or ChatGPT and log in. That’s the whole setup for ClassQuill MCP — no separate app, no extra login for your team to manage. Either way, you’re a sentence away from an answer that used to take a dashboard.

New to the idea of MCP itself? We’ll have a plain-English explainer on what MCP actually is and why a tutoring business should care up soon (link on publish). Curious what this looks like run entirely by hand, without ClassQuill, and why that’s exactly the gap ClassQuill closes? Read how one tutoring company runs on Claude Code and Obsidian today.

For the full technical reference — every tool, every auth method, rate limits — see docs.classquill.com/mcp and docs.classquill.com/connectors.