| name | mcp-selection |
| description | Use when choosing, adding, configuring, or reviewing MCP servers, MCP client config, tool permissions, remote server URLs, local stdio commands, or MCP-related security tradeoffs. |
MCP Selection
Use this skill to add tools deliberately instead of collecting servers.
Selection Flow
- Name the job the MCP server must do.
- Check whether built-in shell, browser, git, or repo tools already cover it.
- Prefer official or high-trust servers with active maintenance.
- Start with read-only or narrow toolsets.
- Add one server at a time and verify it with a small real task.
Risk Review
Before enabling a server, check:
- What data can it read?
- What can it write or delete?
- Does it execute shell commands or arbitrary code?
- Does it call external networks?
- Where are credentials stored?
- Can permissions be narrowed by path, token scope, toolset, or database role?
Config Rules
- Keep committed config as examples only.
- Use placeholders for secrets.
- Prefer local env vars or secret managers for tokens.
- Restrict filesystem roots to project directories.
- Use read-only database credentials by default.
- Document why each server exists.
Output
For each proposed server, include purpose, install/config command, required secrets, permission scope, and why it is worth the added surface area.
