@secr/mcp

Secr MCP server — gives AI coding agents (Claude Code, Cursor, Copilot) direct access to secrets via the Model Context Protocol.

Install

npm install @secr/mcp

claude mcp add secr -e SECR_TOKEN=secr_agent_xxx -- npx @secr/mcp

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "secr": {
      "command": "npx",
      "args": ["@secr/mcp"],
      "env": { "SECR_TOKEN": "secr_agent_xxx" }
    }
  }
}

Add to .vscode/mcp.json:

{
  "servers": {
    "secr": {
      "command": "npx",
      "args": ["@secr/mcp"],
      "env": { "SECR_TOKEN": "secr_agent_xxx" }
    }
  }
}

The server exposes 5 tools to AI agents:

Tool	Description
`get_secret`	Get a single secret value by key
`list_secrets`	List secret key names (no values) with optional search
`set_secret`	Create or update a secret
`delete_secret`	Delete a secret
`list_environments`	List environments for a project

All tools accept optional org, project, and environment parameters. When omitted, they resolve from environment variables or .secr.json.

Variable	Required	Default	Description
`SECR_TOKEN`	Yes	--	Agent token (`secr_agent_...`)
`SECR_ORG`	No	`.secr.json`	Organization slug
`SECR_PROJECT`	No	`.secr.json`	Project slug
`SECR_ENVIRONMENT`	No	`.secr.json`	Default environment slug
`SECR_API_URL`	No	`https://api.secr.dev`	API base URL

secr agents create --name "claude-code" --scope "read:secrets,write:secrets"

Agent tokens use scoped permissions — the server only has access to what the token allows.

Full docs at secr.dev/docs.

MIT