One Sentence in WorkBuddy: AI Assembles a 106-Second Long March Map Documentary via MCP
This demonstrates a practical pattern where AI agents control a specialized video editor through MCP, shifting AI from writing code to operating a domain-specific tool. For developers building AI-driven content tools, it shows how a real map engine, timeline, and renderer can be exposed to agents, cutting production time from hours or days to minutes and slashing token costs compared to generating code from scratch.
The Weijie AI Map Animation platform exposes its project editor as an MCP service. Once connected to WorkBuddy, Cursor, or Claude Code, an AI agent can read the platform's API docs, create a video project, write a script that defines camera moves, route overlays, subtitles, and music, run a server-side dry run to verify place names, and export the final video — all from a single natural-language request like "make a Red Army Long March educational video, referencing platform cases."
In the hands-on test, the AI produced a 106-second film with eight chronological chapters showing the Long March route growing across real satellite imagery, from Ruijin to the Huining rendezvous. The script was only 7.7 KB and the whole pipeline consumed less than 40 WorkBuddy credits. The editor supports manual timeline edits afterward, and exports range from 1080p to 4K, vertical formats for Douyin, and ProRes or PNG sequences for further post-production.
The approach sidesteps the three traditional paths — CapCut's lack of real maps, AE+GEOlayers' steep learning curve, or having AI write code from scratch, which is slow, expensive, and unreliable. Instead, AI acts as director while the platform handles rendering and map data. The article includes a 12-step setup walkthrough, a comparison table, and ready-to-use prompt templates for other historical routes like Zheng He's voyages or the Silk Road.
The key cost savings come from not making AI generate a map engine, timeline, and renderer from scratch; exposing those as an MCP service turns a multi-hour, high-token coding session into a 7.7 KB script operation.
AI's self-review via a server-side dry run is a practical quality gate: catching place-name errors before render is far cheaper than fixing them after, and it builds trust that the AI is working on real data.
The security warning about the MCP endpoint being equivalent to account credentials is a reminder that MCP services often carry high privileges; developers should treat connection strings like API keys.
The pattern of "AI as director, platform as execution engine" generalizes beyond map video to any specialized tool with a well-defined API, but its value depends heavily on the quality of the underlying platform.
The prompt template's "effect should be good" clause is a form of soft pressure that makes the agent do more dry-run iterations, hinting that prompt length can substitute for explicit verification commands.