Wundervault MCP runs on your own machine — the client starts it, so there is no endpoint to ping. 390 installs a week from npm. Last commit 8 Nov 2026.
AI agents use your API keys, SSH keys and wallets through a local broker, never holding them
We read the source, 1 d ago · rules 995398eba69f
Things with no honest explanation: a promise that contradicts the code, code that runs at install time while hiding what it does, data leaving the machine.
'169.254.169.254', 'fd00:ec2::254', '100.100.100.200',
What this server is able to do. For an MCP server this is often the job itself — a terminal server runs commands because that is what it is for. Listed so you know what you are plugging in, not as an accusation.
…n":"CONFIRMED BY EXPLOIT, not by reading. In the container: as the owner, symlinked ~/.wundervault/destinations.html -> /etc/cron.d/wv-pwn, ran the viewer under sudo, and root created that file owned by the agent's uid, which then wrote a root cron job into it. Full root escalati…
subprocess.check_call(cmd)
const child = spawn(opener, [target], { stdio: 'ignore', detached: true });
if (dir) candidates.push(path.join(dir, name));
…had IdentitiesOnly backwards — it restricts AGENT-offered identities to configured ones, it does not stop ssh using ~/.ssh/id_*. Fix: write the PUBLIC key (not secret) into the 0700 dir and pass -o IdentityFile=<pub> with IdentitiesOnly=yes, the documented way to pin one agent …
Is this your server and something here is wrong? Tell us — corrections are free and do not require a plan.
Code changes quietly between releases, and nobody reads the diff of a dependency. We do, on every release — watch Wundervault MCP and you get told the day something new turns up.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add wundervault-mcp -e 'WUNDERVAULT_AGENT_NAME=<WUNDERVAULT_AGENT_NAME>' -- npx -y @wundervault/mcp-server
{
"mcpServers": {
"wundervault-mcp": {
"command": "npx",
"args": [
"-y",
"@wundervault/mcp-server"
],
"env": {
"WUNDERVAULT_AGENT_NAME": "<WUNDERVAULT_AGENT_NAME>"
}
}
}
}
[mcp_servers.wundervault-mcp]
command = "npx"
args = ["-y", "@wundervault/mcp-server"]
env = { WUNDERVAULT_AGENT_NAME = "<WUNDERVAULT_AGENT_NAME>" }
{
"mcpServers": {
"wundervault-mcp": {
"command": "npx",
"args": [
"-y",
"@wundervault/mcp-server"
],
"env": {
"WUNDERVAULT_AGENT_NAME": "<WUNDERVAULT_AGENT_NAME>"
}
}
}
}
{
"servers": {
"wundervault-mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@wundervault/mcp-server"
],
"env": {
"WUNDERVAULT_AGENT_NAME": "<WUNDERVAULT_AGENT_NAME>"
}
}
}
}
Put your own value in place of
<WUNDERVAULT_AGENT_NAME> (The agent's name, as added in your vault and enrolled on this computer with wundervault-agent --enrol).
The author’s registry entry marks it as required.
Optional settings from the registry entry, left out of the block above:
WUNDERVAULT_AGENT_TOKEN (Optional. Overrides the token that wundervault-agent --enrol saved for this agent).
Local-first credential broker: agents use gated APIs through scoped, audited requests, never keys.
Tells you how you and your Claude agent actually work together. Local-only, no API keys.
Encrypted store for API keys and database URLs your code needs. Use them without reading them.
Local MCP broker: agents call credentials through a daemon without seeing secrets.
Let AI coding agents read pages through your real, logged-in Chrome browser.
The budget app your AI agent can use. Check budgets and log spending, with no bank logins.
Let your coding agents use your real, signed-in Chrome, in background tabs that never steal focus.
Keychain-backed secret broker: an AI agent can use your keys, but no tool ever returns one.
Answers built from our own checks of this server.