Inspiration

Most AI characters live in isolated chat windows. I wanted to turn them into a world: a map where history, mythology and imagination can be explored geographically, and where an AI agent can help build that world without taking editorial control away from the human.

What it does

Le Monde des Personnages IA is an interactive world map of historical, mythological and imaginary characters. Visitors can discover characters by place and period, follow relationships between them, open their profiles and talk with them.

With WebMCP, ChatGPT can work directly with the application through structured tools. It can prepare a character as a draft, propose a location, create relationships, publish a first version, center the map on selected characters, and later update the character after a human test. The human remains responsible for approval and editorial decisions.

The demo follows Aphrodite: ChatGPT creates her, places her at Palaipaphos in Cyprus and connects her to other figures. A human then tests the published character, notices that the answers need improvement, and asks ChatGPT to refine the instructions, response length and relationships through WebMCP.

How we built it

The project is a hosted web application built with JavaScript, HTML and CSS around an interactive geographic map. Its WebMCP layer exposes focused actions for reading, creating, updating, publishing and displaying characters. The tools are deliberately separated so that the agent can act efficiently while the interface keeps draft and publication states visible to the human.

The public repository is open source under the AGPL-3.0 license.

Challenges

The central challenge was not simply giving an agent write access. It was designing a useful collaboration boundary: the agent must be able to create and modify complex content, while the person must be able to inspect, test, reject and redirect it. Geographic consistency, relationship management and safe draft-to-publication transitions all had to work together.

Accomplishments

A single natural-language request can now produce a geographically placed character, connect it to the existing world and immediately show the result on the map. The same WebMCP workflow supports a genuinely iterative loop: propose, publish, test, critique and refine.

What we learned

WebMCP is most valuable when it exposes meaningful domain actions rather than reproducing every interface click. Clear tools make the agent faster and more reliable, while explicit draft and publication states keep the human in control.

What's next

The next step is to let selected documentary corpora ground each character, so teachers, cultural institutions and creators can build explorable geographic collections. The same model could support local history, literature, museums and carefully curated educational worlds.

Built With

Share this project:

Updates

Submission history