CodeCraftMCP server · validation & data lookup for Bedrock
https://codecraft-ashy-seven.vercel.app/mcp

Limits & data

WHAT IT DOES NOT CATCH
Valid but not intended
The file is correct, the behaviour is wrong. Example: sending a player a message on the worldLoad event — the event fires, but there is no player there to receive it. No schema can catch this.
Context
An existing Molang query used in the wrong context is rejected by the game. The check only looks at whether the query exists.
Silent failureMEASURED
An entity pointing at a loot table that does not exist loads, spawns and dies. The game writes nothing; nothing drops.
Untried document typesMEASURED
The pack that was verified in-game carried 13 files and touched 11 document types. The number of recognised types is 60.
WHAT WAS MEASURED

This project separates “it works” from “it was measured”.

MEASUREDThe validator’s error findings and the game’s ContentLog errors converged on the same set for one test pack: both empty.
MEASUREDAn unknown Molang query drops the entire block definition in-game. After the measurement, the check was raised to error.
MEASUREDAn unknown component name is rejected the same way, but the check stayed a warning, because the source list had a measured gap of 126 names. The gap has been closed; the severity decision is being handled separately.
A rule is not coded until it is measured, and everything measured is written down with its history.
WHERE THE DATA COMES FROM

Derived facts, not republication. No raw content is served; facts like “this id exists” and “this field is required” are indexed.

Mojang/bedrock-samples
Block, entity and item identifiers; command grammar; Molang queries; component names; schemas
Minecraft EULA — derived index only
Blockception/Minecraft-bedrock-json-schemas
The schemas JSON validation uses
BSD-3-Clause, attribution required
@minecraft/* npm packages
Script type definitions
MIT, Microsoft
MicrosoftDocs/minecraft-creator
Release notes
CC-BY-4.0
How fresh the data is
  • Data indexes update automatically, daily.
  • The scheduled run is set to 05:00 UTC, but it was measured starting about 5 hours late on average. So the data can be at most 1 day + ~5 hours old.
  • Data is foldered by game version, and the tools state which version they used in their answer.
Hosting and privacy
  • Hosted on a free tier. That is a design choice, not a constraint being hidden.
  • There is no rate limit on the endpoint, and that is stated openly.
  • No tool-usage log is kept on the server; the server is stateless.
  • No personal data is collected. There are no accounts.
Source and contributing

The repository is public and licensed Apache-2.0. Issues are the way to report a problem. The measurement logs sit in the repo in the open — claims being checkable is a feature here.