cue/mcps — MCP Server Registry

The MCP server configurations used by cue. Each profile references servers from this registry by ID.

What's here

mcps/
├── configs/              Sanitized JSON configs (no secrets)
│   ├── claude.sanitized.json
│   ├── claude_runtime.sanitized.json
│   └── codex.sanitized.json
├── mcps/                 Per-server documentation
│   ├── gbrain/           Personal wiki with embeddings
│   ├── colony/           Multi-agent coordination
│   ├── coolify/          Deployment management
│   ├── higgsfield/       AI image/video generation
│   └── ...               13+ servers
├── plugins/              Claude Code plugin configs
├── scripts/              Sync and maintenance scripts
└── templates/            New MCP server templates

How cue uses this

When you run cue use backend, cue reads configs/claude.sanitized.json to find the MCP server entries for that profile's declared MCPs. The materializer writes them into the runtime's settings.json.

# profiles/backend/profile.yaml
mcps:
  - coolify        # ← looked up in configs/claude.sanitized.json
  - gbrain         # ← inherited from core

Adding a new MCP

1. Add the server config to configs/claude.sanitized.json:

   "my-server": {
     "command": "npx",
     "args": ["my-mcp-server"]
   }

2. Create docs at mcps/my-server/skills.md describing what tools it exposes.

3. Reference it in a profile: mcps: [my-server]

Security

Configs are sanitized — no API keys, tokens, or secrets. Environment variables use <redacted> placeholders. Real values live in ~/.claude.json (never committed).

Related

• cue — the profile manager that consumes this registry
• resources/skills — the skill library