truenas-mcp

An MCP server for TrueNAS SCALE. Read-first, deployable as a TrueNAS app, and authenticated with each user's own API key.

Status: working, young. Every capability in the design is implemented and verified against a live TrueNAS 26 box. Expect rough edges rather than gaps.

Why this exists

iX ship an official truenas/truenas-mcp, and if it fits your needs you should use it. This one exists for three things it does not do:

• It only speaks stdio, so it cannot be deployed as a container on the NAS and reached from elsewhere.
• It holds a single server-wide API key, so every caller gets identical reach no matter how they authenticate.
• Its app coverage is catalog-shaped — install, uninstall, browse — with no way to pull new images and redeploy an app you already run.

Design

Three ideas do most of the work.

The credential is the authorization. Callers supply their own TrueNAS API key; this server stores none. A session reaches exactly what that user's key permits, revocation happens in the TrueNAS UI, and there is no shared secret to leak. Authentication and authorization stop being two systems that can disagree. Served over stdio the key arrives from the environment instead, because there is no request to carry it — but the client spawns one process per user, so it is still that user's own key. What the design rules out is one key standing in for many callers, not configuration as such.

Reads and writes get different tool shapes. Reads are grouped into concern-level tools with an op enum, because they share most of their arguments and 815 middleware methods cannot each become a tool. Writes are individual tools — MCP annotations are per-tool, so bundling a safe operation with a destructive one behind one op parameter would put both behind a single consent gate, and a user who tires of confirming list_pools will allowlist the tool that can also export a pool.

Read-only by default. Mutating tools appear only when explicitly enabled. Separately, a denylist of unrecoverable operations is not reachable under any configuration, and it constrains argument values rather than just method names — deleting an app is recoverable, deleting it along with its volumes is not, and those are the same method.

Each of these was measured against a live box rather than reasoned about in the abstract, and several were overturned by what that measurement found.

Requirements

• TrueNAS SCALE 25.04 or later. The REST API is removed in TrueNAS 26; this server speaks only the versioned JSON-RPC 2.0 WebSocket API.
• A TrueNAS API key per user. Create them under Credentials → API Keys.

Deploying as a TrueNAS app

Copy deploy/truenas-custom-app.yaml, adjust TRUENAS_MCP_TARGET, and paste it into Apps → Discover → Install via YAML.

It mounts no host socket and requests no privileged access. The server reaches the middleware over the network even when running on the same box, so that every connection carries a user identity rather than root-equivalent socket access.

Running it elsewhere

Nothing requires the server to run on the machine it manages, and there is a good reason not to: installed as a TrueNAS app, it is unavailable exactly when the box is unhealthy — which is when you most want to ask it what is wrong.

docker run -p 8080:8080 \
  -e TRUENAS_MCP_TARGET=nas.local \
  -e TRUENAS_MCP_TLS_CERT=/tls/cert.pem \
  -e TRUENAS_MCP_TLS_KEY=/tls/key.pem \
  ghcr.io/cedricziel/truenas-mcp:main

Running the binary

Each GitHub release attaches binaries for Linux, macOS, and Windows on amd64 and arm64, alongside a checksums.txt. Configuration is environment variables only — there is no config file, and the only flags are --stdio and --healthcheck.

By default the binary is an HTTP server. Running it does not make a client pick it up on its own; it listens on a port, and the client connects to it by URL. For clients that spawn the server themselves, see Serving over stdio below.

TRUENAS_MCP_TARGET=nas.local \
TRUENAS_MCP_LISTEN=127.0.0.1:8080 \
TRUENAS_MCP_TARGET_INSECURE=true \
TRUENAS_MCP_ALLOW_PLAINTEXT=true \
./truenas-mcp

Point the client at http://localhost:8080/mcp, with the TrueNAS API key sent as an Authorization: Bearer header, the same as in Connecting a client.

TRUENAS_MCP_TARGET_INSECURE is typically needed for the reason given in On the two TLS settings: TrueNAS ships a self-signed certificate for CN=localhost that will not validate against any other address. TRUENAS_MCP_ALLOW_PLAINTEXT is defensible here specifically because TRUENAS_MCP_LISTEN binds the listener to loopback — reachable only from the same machine — which is the condition that section argues plaintext requires. The default bind address is :8080, which is every interface, so dropping that setting while keeping plaintext would put API keys on the wire.

Serving over stdio

Some clients spawn a server as a subprocess and talk to it over its standard input and output rather than connecting to a URL. --stdio serves the same tools that way.

claude mcp add --scope user truenas \
  --env TRUENAS_MCP_TARGET=nas.local \
  --env TRUENAS_MCP_TARGET_INSECURE=true \
  --env TRUENAS_MCP_API_KEY=$YOUR_TRUENAS_API_KEY \
  -- /path/to/truenas-mcp --stdio

There is no request to carry a header here, so the key comes from TRUENAS_MCP_API_KEY instead. That is not the shared secret the HTTP transport avoids: the client spawns one process per user, so the key it passes is that user's own, and the process reaches exactly what that key permits. The same variable is refused in HTTP mode, where one process serves many callers and a configured key would be shared by all of them.

Nothing about the listener applies. TRUENAS_MCP_LISTEN, the two TLS settings and TRUENAS_MCP_ALLOW_PLAINTEXT are ignored with a warning rather than an error, since none of them weakens anything when no listener exists. --healthcheck is refused alongside --stdio, because it probes a listener that was never started.

Settings that concern the target rather than the listener still apply, including TRUENAS_MCP_TARGET_INSECURE and TRUENAS_MCP_ENABLE_WRITES.

Configuration

All configuration is environment variables; no config file or persistent volume is needed. Invalid configuration refuses to start rather than running degraded.

No credential is configurable for the HTTP transport. Callers supply their own with each request, and setting TRUENAS_MCP_API_KEY without --stdio is a startup error rather than a silent fallback. Over stdio there is no request to carry one and the process serves a single user, so the variable is how that user's key arrives — see Serving over stdio.

On the two TLS settings

Transport scheme and certificate verification are deliberately separate.

TrueNAS ships a self-signed certificate issued for CN=localhost with only DNS:localhost as a SAN, so no address you can reach it by will validate. The fix is TRUENAS_MCP_TARGET_INSECURE=true, which keeps the connection encrypted and merely unauthenticated. If certificate problems forced you onto plaintext instead, TrueNAS would see your API key in the clear — and revoke it.

TRUENAS_MCP_ALLOW_PLAINTEXT is about the boundary callers cross, which carries their API keys. It is correct when a reverse proxy terminates TLS in front of the server, and wrong when the plaintext listener is reachable directly.

Connecting a client

claude mcp add --scope user --transport http truenas \
  https://your-host/mcp \
  --header "Authorization: Bearer $YOUR_TRUENAS_API_KEY"

The key may also be sent as X-TrueNAS-API-Key, for clients that cannot set an Authorization header. Requests without either are refused with 401.

Clients that spawn the server rather than connect to one want Serving over stdio instead.

Current state

Working:

• Streamable HTTP transport, per-session credentials, 401 without one
• Stdio transport under a single per-process credential, for clients that spawn the server rather than connect to one
• JSON-RPC middleware client: concurrent calls on one connection, structured errors distinguishing unreachable / unauthenticated / unauthorized / rate limited, and interrupted requests reported as may have been applied
• Session reconnection when a connection dies, and refusal to run against a release older than 25.04
• Container image, CI, GHCR publication, TrueNAS app deployment

Not implemented: job progress via resource subscription. Polling covers the same ground and is the path the design treats as reliable — subscription was always an enhancement over it, and MCP client support for it is thin.

Tools