MCP server for AI agents

Let agents hand off secrets securely — encryption runs locally in the MCP process.

Who this is for

Use MCP when an AI assistant needs to create a PrivateNote link as part of a workflow — for example, sending an API key securely with burn-after-reading.

The MCP server exposes a create_private_note tool. The local process encrypts content and returns a shareable URL to the agent.

Install and configure

# npm package (runs via npx)
npx -y privatenote-mcp

# Cursor — ~/.cursor/mcp.json
{
  "mcpServers": {
    "privatenote": {
      "command": "npx",
      "args": ["-y", "privatenote-mcp"]
    }
  }
}

# Claude Desktop — claude_desktop_config.json (same structure)

Try it live

Simulate an MCP tool call — the playground encrypts in-page and returns the JSON shape agents receive from create_private_note.

Live demo — creates a real note on privatenote.ai. Use a throwaway secret, not production credentials.

Terminal
// MCP tool call
create_private_note({ "content": "sk-live-demo-replace-me", "expiresIn": "1h", "burnAfterReading": true })

Security architecture and privacy boundary

Encryption happens in the local MCP process before upload. The server stores ciphertext only.

Important: if you paste a secret into a cloud AI prompt, the model provider can see it before the MCP tool runs. For the strongest posture, use a local model or the VS Code extension for secrets the agent must never see.

Encrypt locally

AES-256-GCM on your device — browser, editor, terminal, or MCP process.

Upload ciphertext

Only encrypted bytes and note metadata (expiry, burn-after-reading) reach the server.

Key in #fragment

The decryption key stays in the URL hash. Browsers do not send fragments in HTTP requests.

privatenote.ai/note/…#key

Full diagram on developer hub
  • creationSource: mcp in analytics.

  • Optional PRIVATENOTE_API_BASE_URL for self-hosted instances.

  • Works with any MCP-compatible client.

Configuration reference

NameDescription
create_private_noteMCP tool — content, expiry, burn-after-reading
commandnpx (or node path to local build)
args["-y", "privatenote-mcp"]
PRIVATENOTE_API_BASE_URLOverride API for self-hosted deployments

Related integrations

MCP server for AI agents

Let agents hand off secrets securely — encryption runs locally in the MCP process.