Improve MCP tool descriptions and feedback for higher quality builds#2
Improve MCP tool descriptions and feedback for higher quality builds#2
Conversation
- Enrich execute_commands tool description with block state syntax (doors, stairs, slabs, logs, etc.), fill command options, and building best practices to guide LLM agents - Improve get_player_info description with clear field explanations and guidance to use absolute coordinates - Improve get_blocks_in_area description to encourage post-build verification workflow - Increase default maxAreaSize from 10 to 48 so agents can verify larger structures in a single scan - Return structured per-command JSON results from execute_commands instead of plain text, including success/failure per command and a hint to verify with get_blocks_in_area https://claude.ai/code/session_01H99Nfmq8L9UkwYapriA4Bx
- Fix bed placement: head goes in facing direction (south=+Z), not -Z - Remove banner from glazed terracotta line (standing banners use rotation 0-15, not facing) - Fix max fill volume: 32768 (Minecraft limit), not 125000 (mod limit) https://claude.ai/code/session_01H99Nfmq8L9UkwYapriA4Bx
|
@codex review this PR. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7942ada8d6
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
| executeCommandsTool.addProperty("description", | ||
| "Execute one or more Minecraft commands sequentially. " + | ||
| "Allowed commands: fill, clone, setblock, summon, tp, give, gamemode, effect, enchant, weather, time, say, tell, title.\n\n" + |
There was a problem hiding this comment.
The execute_commands description now hard-codes an allowlist, but enforcement is runtime-configurable (SafetyValidator.validate checks config.getServer().getAllowedCommands()). In environments that customize server.allowedCommands, this metadata becomes inaccurate and agents will repeatedly issue commands that are rejected (or miss commands that were explicitly enabled), which hurts tool-call reliability. Build this list from the loaded config so the advertised contract matches actual validation.
Useful? React with 👍 / 👎.
2 participants
(doors, stairs, slabs, logs, etc.), fill command options, and
building best practices to guide LLM agents
and guidance to use absolute coordinates
verification workflow
larger structures in a single scan
instead of plain text, including success/failure per command and
a hint to verify with get_blocks_in_area
https://claude.ai/code/session_01H99Nfmq8L9UkwYapriA4Bx