Build an AI exam tutor in 10 minutes with MirrorMingo + MCP
In our launch post we opened up verified African exam content via a REST API and six SDKs. This post is about the other way in: the Model Context Protocol (MCP) server, which lets an AI agent use MirrorMingo as a native tool.
Why MCP matters for edtech
A lot of new study software isn't a traditional app at all — it's an agent. A chat tutor, an IDE-style "explain this" assistant, a WhatsApp study buddy. Those agents are only as good as the tools they can reach.
The problem with most AI tutors is the same one we've always had: when an LLM invents exam questions, students practise on material that won't appear and sometimes isn't even correct. MCP fixes the plumbing; MirrorMingo fixes the content. Together, the agent stops guessing and starts pulling real past papers, then marks the student against the verified answer key.
What is MCP, briefly
The Model Context Protocol is an open standard for connecting AI assistants to tools and data. A "server" exposes tools; a "client" (Claude Desktop, Cursor, Claude Code, …) lets the model call them. MirrorMingo ships an MCP server, so any of those clients can use our exam tools with zero glue code.
Setup
pip install mirrormingo
export MIRRORMINGO_API_KEY=mirrormingo_xxx # from Account → Developer
Add it to Claude Desktop's claude_desktop_config.json:
{
"mcpServers": {
"mirrormingo": {
"command": "mirrormingo-mcp",
"env": { "MIRRORMINGO_API_KEY": "mirrormingo_xxx" }
}
}
}
Restart Claude Desktop. That's the whole integration.
Use it
You: Use MirrorMingo to give me 5 real JAMB Physics questions, one at a time. Mark each answer and explain what I get wrong.
Behind the scenes the agent calls list_mock_papers to find a Physics paper,
open_mock_paper to pull real questions, and submit_mock_attempt to score
your answers — all against verified content, none of it invented.
The tools
| Tool | What the agent does with it |
|---|---|
list_exams | Discover supported exams |
get_exam_overview | Understand structure and scoring |
list_mock_papers | Find a real past paper |
open_mock_paper | Start a timed attempt |
submit_mock_attempt | Score answers with explanations |
Safe by design
The MCP server is a thin layer over the same public API as our SDKs. It uses your API key, so it inherits the same authentication and rate limits — an agent can never reach beyond what your key allows. And because it's the same contract, your MCP tutor and your REST-based app stay perfectly in sync.
SDK or MCP — which?
Both, usually. Use the SDKs for deterministic app logic (a timed exam UI, a nightly sync job). Use the MCP server when an LLM should decide when to fetch questions and how to coach. Same content underneath.