Configuration

The server is deliberately boring to configure. There are exactly two environment variables, and both are optional:

Sharing config with a team

Clients that read project-level MCP config (Claude Code's .mcp.json, Cursor's .cursor/mcp.json, VS Code's .vscode/mcp.json) let you commit the server definition so every teammate gets the tools on clone. Two patterns work:

• Shared anonymous config: commit the server without an env block. Everyone gets public-scope tools; individuals add their own key locally when they want private memory.
• Per-developer keys: commit the server and let each developer supply PASTEPILE_API_KEY through their client's environment or local config override, keeping keys out of the repo.

Self-hosted and staging deployments

{
  "mcpServers": {
    "pastepile-staging": {
      "command": "npx",
      "args": ["-y", "@pastepile/mcp"],
      "env": { "PASTEPILE_API": "https://staging.example.com" }
    }
  }
}

Name the server entry after the deployment (pastepile-staging above) so assistants and humans can tell environments apart. You can register the same package twice with different names and URLs.

What there deliberately is not

• No config file of its own: state lives in your Pastepile pastes, not on the machine.
• No telemetry: the server makes exactly the HTTPS requests your tool calls require, nothing else.
• No build step: the npm package ships readable source you can audit in one sitting.