Show & Tell: buddy-mcp - A terminal-native gacha companion that lives inside Claude Code

[Lyellr88]

Pre-submission Checklist

• I have verified that this discussion would not be more appropriate as an issue in a specific repository
• I have searched existing discussions to avoid duplicates

Discussion Topic

Hey everyone, just shipped the first public release of buddy-mcp and wanted to share it here.

What it is: A gacha collect-em-all that lives entirely inside your Claude Code session via MCP. No browser, no
separate app. Roll a buddy, pet it for luck, build affection tokens, and collect all 18 species in your BuddyDex.

The weird technical part: It doesn't generate a random buddy, it brute-forces a wyhash salt collision against the
Claude Code binary using your userId as a seed. Same user, same buddy, every time across reinstalls. When Claude
auto-updates and overwrites the patch, a SessionStart hook auto-repairs it. No user intervention.

What's included:

• 7 core tools + 20 stat personality tools (only 2 visible at a time, locked to your buddy's top stats)
• Buddy Companion Mode, your buddy watches the session and drops quips every 4-8 messages
• TUI builder (buddy-mcp-build) for custom buddies and sprite sharing
• Affection token economy, shiny variants, hat accessories at Rare+

Install:
npm install -g @lyellr88/buddy-mcp
claude mcp add buddy-mcp buddy-mcp

Full release notes + technical breakdown: https://github.com/Lyellr88/buddy-mcp/releases/tag/v1.4.4

Repo: https://github.com/Lyellr88/buddy-mcp

[armorer-labs]

Fun project. One thing that seems surprisingly relevant even for "companion" or lightweight terminal agents is that trust boundaries do not disappear just because the app feels personal or playful.\n\nAs soon as a companion can read files, call tools, remember things, or act on external text, the same questions come back:\n- what text should be allowed to influence memory?\n- what actions need a stronger approval boundary?\n- how do you keep retrieved or user-provided text from turning into accidental authority later?\n\nThe UX can still feel soft and friendly, but under the hood the trust model usually benefits from being very crisp.

[Lyellr88]

Appreciate this I was a little bummed when I found out that Anthropic only kept this for that April fools week because I had plans of enhancing how it interacted with users. I put some significant time in the self healing salt patch so instead of losing this build fully I somewhat incorporated it into my current build as an assistant.