Now You Can Ask an AI Assistant to Search ImpactMojo

The idea in one sentence

If you have used an AI assistant recently, you have probably run into its blind spot: it knows a great deal in general, but it does not know your particular tools, your datasets, or the specific body of knowledge you work with. The Model Context Protocol, usually shortened to MCP, is a shared standard that closes that gap. In plain terms, it is a standard way to let an AI assistant look things up in a specific library. A useful mental picture is a USB port for AI: plug in a connection, and the assistant can reach a particular set of information and answer from it.

We have built one of these connections for ImpactMojo. With it in place, an AI assistant no longer has to guess about a behaviour change technique or paraphrase a half-remembered course description. It can look the answer up in our actual material and hand it back with the real details.

[Illustration 1: ImpactMojo MCP Server architecture]

What you can now ask

Once the connection is in place, an AI assistant can reach into most of what ImpactMojo holds and answer precisely. That includes:

• A search across more than 1,300 indexed items (courses, games, labs, handouts, book summaries, behaviour change techniques, Dataverse entries and more).
• All 203 behaviour change techniques, each with South Asian context and 50 of them with case studies, so the assistant can pull up the right one for a given problem.
• The Dataverse: our catalogue of 335 tools, datasets and data sources, browsable by category, such as Climate & Environment or Health & Epidemiology.
• Practice challenges and their full case material, filtered by learning track and difficulty.
• The economics simulation games that have AI characters, with the concept each game teaches and each character’s backstory.
• India's greenhouse-gas emissions data, drawn from the Climate TRACE project.

Put simply, the assistant stops answering from memory and starts answering from the library.

[Illustration 2: The kinds of things an AI assistant can look up in ImpactMojo]

Why this matters

No practitioner should have to hold more than 1,300 indexed resources in their head, or remember which of 203 behaviour change techniques speaks to social support, or which dataset maps poverty from satellite imagery. The point of connecting the library to an AI assistant is that the right course, technique or dataset can surface at the moment it is actually needed.

We think development education is heading this way: knowledge that meets practitioners inside the tools they already use, so they do not have to remember to visit a platform and find their way around it.

"The best educational platform is the one that shows up with the right resource at the right moment. This connection is a step toward that."

How to try it

If you use an AI assistant that supports these connections (Claude Code and Claude Desktop are two), you can add ImpactMojo by downloading the code, building it with Node.js and adding one entry to the assistant’s settings. The server runs on your own computer, reads the data files that ship with it, makes no outside calls and needs no account or key. Step-by-step instructions for both are in our open-source project on GitHub. The code is under the MIT licence. The course and handout content it returns is under CC BY-NC-ND 4.0, so check the licence before you redistribute it.

What's next

This is an early version. We plan to connect more of the library over time, including our reading companions and dojo practice sessions, and to document more AI assistants than the two covered today.

The code is open source, and we would like to hear from anyone who builds something with it. If you are thinking about what AI tools mean for evaluation practice, our free course AI for Impact looks at it from the evaluator’s side.