Short answer
Do not put passwords or private tokens in MCP settings files. Give each server only the access it needs. Write down the private settings people must add, test the connection, and review shared files in Git. Package only files that are safe to share.
Words used in this guide
- AIPM: a tool that installs and updates reusable AI instructions.
- Package: a named group of files that people can install and update together.
- Repository or repo: a project folder whose changes are saved and tracked.
- Git: a tool that records file changes so people can review or undo them.
- Config: settings that tell a tool how to work.
- MCP: a standard way for an AI tool to connect to other tools and information.
- Environment variable: a private setting stored on your computer, often used for a password or token.
What this means
Keep secret values local
Config can name the environment variables a server needs, but it should not contain real tokens or passwords. Use local environment settings or a secret manager for values.
Limit access and explain the purpose
Add only the servers a project needs. Explain what each server can see and when the AI should use it. Also explain which actions need approval from a person.
Make setup repeatable
Store safe config and test steps in Git. If many repos need the same server, package the public setup with AIPM while each user supplies private values locally.
Simple steps to create and AI agent
- Choose the smallest server access that completes the job.
- Move tokens and passwords into environment variables.
- Document the server purpose and required variables.
- Test a safe read action before allowing write actions.
- Package only public config and setup instructions.
Where to go next
If you want to try this in a real project, start with the install guide, then read how to use AIPM. If you want to share your own workflow, read the publishing guide.
FAQ
Should mcp.json contain API keys?
No. Refer to environment variable names and keep the real values outside files that can be committed or published.
Can MCP configuration be shared across repos?
Yes. Share safe server definitions, setup notes, and tests. Keep secrets and repo-specific access choices local.