frankxai MCP Governance
MCP is the agent boundary layer. Use it when an assistant needs typed, authenticated access to a bounded capability.
Design Rules
- Expose tools, not whole systems.
- Prefer read or run scopes before management scopes.
- Give every tool a narrow name, schema, and expected output.
- Separate test and production tools.
- Fail closed on transport, auth, schema, and parsing errors.
- Treat tool descriptions as untrusted surface; validate behavior in code.
Make MCP
Use Make MCP to let agents run activated on-demand scenarios with explicit inputs and outputs. Keep management scopes off until the scenario set is mature.
n8n MCP
Use n8n MCP to expose selected workflows or to let n8n agents call external MCP tools. Enable only the workflows that should be discoverable.
Tool Boundary Checklist
Return:
- tool name
- user-visible purpose
- input schema
- output schema
- auth method
- scope
- failure mode
- human gate
- audit trail
- revocation path
Human Gates
Require explicit approval before tools send messages, post publicly, spend, delete, invite, change permissions, or promote production deploys.
Validation
When available, run:
python scripts/validate_mcp_config.py path/to/config.json