MCP Configs
Use MCP config to expose approved external tools and context to coding agents. Treat every MCP server as a capability expansion, not just a convenience toggle.
Core Team Policy
- approve read-only servers first
- store secrets outside repository config when possible
- document a human owner for each server
- define whether OAuth-based servers are allowed
- review rate limits, side effects, and data access before rollout
Officially Documented Config Locations
OpenAI Codex
Official docs say Codex stores MCP configuration in:
~/.codex/config.toml.codex/config.toml
The docs also confirm CLI management commands such as:
codex mcp addcodex mcp listcodex mcp login
Claude Code
Anthropic's settings docs list MCP server configuration at:
~/.claude.json.mcp.json
Project settings and personal settings are separate, and Windows ~/.claude paths resolve under %USERPROFILE%\\.claude.
Cline
Official Cline docs confirm MCP config at:
~/.cline/mcp.json
Project-level .cline/mcp.json is also documented in the CLI reference.
Example Approval Checklist
Before approving a server, answer:
- what data can this server read?
- what actions can it take?
- does it mutate production or external systems?
- where are credentials stored?
- who owns break/fix support?
- is the server safe enough for default enablement?
Practical Server Categories
Safer early candidates:
- docs lookup
- issue tracker read access
- read-only filesystem helpers
Higher-risk candidates:
- database write access
- deployment systems
- ticket automation
- cloud resource creation
Cross-Tool Recommendation
Do not try to keep every tool's MCP config identical if the products use different formats. Instead standardize:
- approved server list
- approved auth method
- approved scopes
- naming convention
- owner and rotation policy
Then translate that policy into each tool's config format.