What is the Model Context Protocol?
If you have used an AI assistant recently, you have probably noticed a limitation: the AI knows a lot in general, but it does not know about your specific tools, datasets, or organisational knowledge. The Model Context Protocol (MCP) is an open standard that solves this problem. Think of it as a USB port for AI — plug in an MCP server, and your AI assistant gains structured access to external data sources and tools.
For development education, this is transformative. Instead of an AI assistant guessing about behaviour change techniques or paraphrasing half-remembered course descriptions, it can query a structured knowledge base and return precise, contextual information. That is exactly what the ImpactMojo MCP Server does.
What Our MCP Server Offers
The ImpactMojo MCP Server exposes 11 tools and 3 resources across four domains, giving any MCP-compatible AI assistant full access to our knowledge base.
Search & Discovery
- search_content — Full-text search across 700+ content items (courses, games, labs, handouts)
- browse_dataverse — Browse 270 tools, datasets, APIs, and MCP servers by category
- search_dataverse — Keyword search across all Dataverse items
Behavior Change Techniques
- lookup_bct — Get full technique details for any of 203 BCTs, with South Asian context and case studies
- search_bcts — Search and filter by keyword, sector, category, or evidence level
- list_bct_categories — List all 26 BCT categories with technique counts
Learning & Practice
- list_courses — 9 flagship courses across 6 learning tracks, filterable by track, type, and level
- list_challenges — Practice challenges filterable by track and difficulty
- get_challenge — Full challenge details with case context, rubric, and resources
Simulations & Data
- get_game_info — 15 economics simulation games with AI agent personas
- query_climate_data — India's GHG emissions data from Climate TRACE
The server also exposes three resources: a platform overview (impactmojo://overview), full content catalog (impactmojo://catalog), and learning track descriptions (impactmojo://tracks).
How It Works
The architecture is deliberately simple. The server is written in TypeScript using the @modelcontextprotocol/sdk library. At startup, it loads six JSON data files from the ImpactMojo repository — the search index, BCT repository, Dataverse catalog, challenge bank, game agents registry, and Climate TRACE data — totalling roughly 750KB. All data is held in memory for instant queries.
There are no external network calls. No database connections. No API keys required. The server communicates over stdio transport, which means it runs as a local process alongside your AI client. When your AI asks a question, the MCP server queries its in-memory data and returns structured JSON in milliseconds.
The package is published as @varnasr/impactmojo-mcp-server on GitHub Packages and is MIT licensed.
Getting Started
Connecting the ImpactMojo MCP Server to your AI client takes under a minute.
With Claude Code
claude mcp add impactmojo -- node /path/to/mcp-server/dist/index.jsWith Claude Desktop
Add to your claude_desktop_config.json:
{
"mcpServers": {
"impactmojo": {
"command": "node",
"args": ["/path/to/mcp-server/dist/index.js"]
}
}
}From Source
git clone https://github.com/Varnasr/ImpactMojo.git
cd ImpactMojo/mcp-server
npm install
npm run buildYou can test the server with the MCP Inspector: npx @modelcontextprotocol/inspector node dist/index.js
What You Can Ask
Once connected, here are the kinds of questions your AI assistant can now answer precisely, drawing from ImpactMojo's structured data:
Example Prompts
- "Search for courses about randomised controlled trials"
- "Look up BCT technique 4.1 and show me South Asian case studies"
- "What tools in the Dataverse are relevant to climate adaptation?"
- "List practice challenges for the MEAL track at intermediate difficulty"
- "Show me India's GHG emissions data from Climate TRACE"
- "What economics simulation games cover market failures?"
Why This Matters
Development practitioners should not need to memorise 700+ resources across six learning tracks. They should not need to remember which of 203 behaviour change techniques addresses social support, or which Dataverse tool handles geospatial poverty mapping.
AI assistants with MCP access can surface the right course, technique, or dataset at the moment it is needed. This is the future of development education: structured knowledge bases that meet practitioners where they are — in their AI workflows, not on a platform they have to navigate separately.
"The best educational platform is the one that shows up with the right resource at the right moment. MCP makes that possible."
What's Next
This is version 1.0. We are already planning deeper integration with Supabase for personalised learning progress, more data sources including our 27 book companions and 56 dojo practice sessions, and support for MCP clients beyond Claude. The knowledge base will grow, and the server will grow with it.
The ImpactMojo MCP Server is open source, MIT licensed, and accepting contributions. If you build something interesting with it, we want to hear about it.