The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Volta MCP Server listing page.
MCP server for Volta Notes — create and read burn-after-read encrypted notes from any AI agent.
Notes are end-to-end encrypted using AES-256-GCM. The decryption key lives only in the URL fragment — it is never sent to any server. Notes are stored on the Internet Computer and permanently destroyed after a single read.
AI agents regularly need sensitive information at runtime — API keys, passwords, credentials. Today, users paste these into chat where they're stored permanently in conversation history.
With this MCP server, the pattern becomes:
read_volta_note — secret returned, note permanently destroyedOr in reverse — an agent can use create_volta_note to send credentials to a user via a self-destructing link.
Step 1 — Install globally:
Step 2 — Register the server:
That's it. Restart Claude Code and the create_volta_note and read_volta_note tools will be available.
Why
claude mcp addinstead of editing config files? Claude Code reads MCP servers from its own registry, not from~/.claude/mcp.json. Using the CLI ensures the server is registered correctly. The-s userflag makes it available across all projects.
nodenot found? Use the full path: replacenodewith the output ofwhich node(e.g./usr/local/bin/node).
Add to your claude_desktop_config.json:
The npx config above works with any client that supports the standard command/args MCP format — including Cursor, Windsurf, Cline, Continue.dev, and others. Check your client's MCP documentation for where to add server config.
Volta MCP works with Hermes Agent via its MCP integration. Add the server to your Hermes config (~/.hermes/config.yaml):
Restart Hermes and the create_volta_note and read_volta_note tools will be available to your agent.
Agent-to-agent use case: Hermes agents running autonomously often need to pass credentials between workflows without exposing them in logs or memory. Volta Notes is purpose-built for this — one agent creates a note, passes the one-time URL to the next, and the secret is destroyed on read. Nothing persists in conversation history or tool output logs.
See Hermes MCP documentation for full configuration options.
The $(npm root -g) syntax doesn't work in PowerShell or CMD. Use this instead:
Or use the Claude Desktop / npx method above, which works cross-platform.
create_volta_noteCreates an encrypted note and returns a one-time URL.
| Parameter | Type | Description |
|---|---|---|
content | string | Secret content to encrypt (max 2 KB) |
Returns: A voltanotes.com URL. The recipient opens it once, reads the content, and it's gone forever.
read_volta_noteReads and permanently destroys a Volta note.
| Parameter | Type | Description |
|---|---|---|
url | string | Full Volta URL including # fragment |
Returns: The decrypted note content. The note is permanently deleted from the canister — a second read will fail.
Add this to any agent's system prompt to enable secure credential handoff:
#...) — browsers and servers never transmit fragmentsYou should see volta: ... ✓ Connected. If not, see below.
Did you use claude mcp add? Editing ~/.claude/mcp.json manually won't work — Claude Code reads servers from its own registry. Always use claude mcp add to register servers.
Is node on your PATH? Claude Code's shell has a minimal PATH. If node isn't found, use the full path:
Restart required. After adding or changing an MCP server, fully restart Claude Code (quit and reopen).
The server logs Volta MCP server started to stderr on successful startup. Run claude mcp list to check connection status.
MIT — Unprompted Labs