Skip to main content
Glama

Talos opinionated MCP Server

NOTE: This project is under active development.

Motivation

Modern infrastructure often uses tools such as Terraform/OpenTofu, Ansible, and GitOps. These tools deploy and configure Talos Linux nodes and Kubernetes clusters.

The Talos MCP Server does not replace these tools. It gives AI assistants and automation agents access to information about your Talos clusters.

The default tools are read-only. An optional action tool can be enabled with --allow-destructive to reboot a node.

Instead of running many talosctl commands across nodes, you can ask an AI assistant to collect and analyze the required information. Your existing automation tools remain the main source for infrastructure changes.

Related MCP server: rancher-mcp-server

Overview

The server connects an MCP client, such as an AI assistant, to the Talos API (apid) and, optionally, to the Talos discovery service. It supports multiple Talos clusters and returns structured data.

You can use it to:

  • inspect clusters, nodes, and their Talos and Kubernetes versions

  • read Talos service logs and kernel logs (dmesg) of a node

  • review recent Talos runtime events

  • list cluster members from the discovery service, even when the Talos API is down

  • collect information before you make infrastructure changes

Keep your AI on the leash.

Talos tools

The MCP server provides the following tools:

Tool

Arguments

Description

talos_clusters_list

None

List the configured clusters with their endpoints, credential role, and the tools usable on each. Makes no network calls.

talos_clusters_describe

cluster (optional)

Show every node of a cluster with its role and Talos and kubelet versions.

talos_clusters_event

cluster, node, since, limit, actor_id (all optional)

List recent Talos runtime events, newest first. Defaults: last 1h, 50 events.

talos_clusters_members

cluster, role (optional)

List cluster members as the discovery service sees them: node ID, hostname, role, addresses, and KubeSpan data. Available only on clusters with discovery keys.

talos_node_describe

cluster, node, logs, log_lines, events_since, event_limit (all optional)

Show one node in detail: versions, stage and readiness, resource usage (CPU, load, memory, disks, top processes), every service's state and health, recent events, and the last log lines of the unhealthy services. Secrets in the output are masked.

talos_node_logs

service, cluster, node, kubernetes, tail, grep (optional except service)

Show the last lines of a Talos service's logs (kubelet, etcd, apid, machined, ...) or of a Kubernetes container.

talos_node_dmesg

cluster, node, tail, grep (all optional)

Show the last lines of a node's kernel log. Secrets in the output are masked.

cluster is a talosconfig context name. When it is omitted, the current context is used. Use talos_clusters_list first to find the cluster names. MCP clients can discover the full input and output schemas. You can also list the tools from the command line:

talos-mcp tools --talosconfig /absolute/path/to/talosconfig

When --allow-destructive is enabled and the cluster's credential has the os:operator role, talos_node_reboot accepts cluster, node, and an optional mode (default or powercycle). It returns once Talos accepts the request; it does not wait for the node to come back.

The tools available on each cluster depend on the role of its client certificate: os:reader gets the read-only tools, os:operator also gets the reboot tool.

Installation

For a local installation with Homebrew, run:

brew install sergelogvinov/tap/talos-mcp

You do not need a local installation when you use an MCP server hosted on another machine.

Configure Talos clusters

The server reads a standard talosconfig, the same file talosctl uses. Each context is one cluster. Create a dedicated credential with the smallest role you need:

talosctl config new --roles os:reader --crt-ttl 8760h ~/.talos/mcp-config

A context may also have an optional discovery block with the cluster ID and secret, which enables talos_clusters_members:

context: prod-eu
contexts:
  prod-eu:
    endpoints: [10.0.0.10, 10.0.0.11, 10.0.0.12]
    ca: LS0t...
    crt: LS0t...
    key: LS0t...
    discovery:
      cluster_id: 3x9y...
      cluster_secret: c2VjcmV0...

The private key and cluster_secret can be stored encrypted with a passphrase, an age key, or your SSH key:

talos-mcp config encrypt --talosconfig ~/.talos/mcp-config -o ~/.talos/mcp-config.enc \
    --recipient-file ~/.ssh/id_ed25519.pub
talos-mcp config check --talosconfig ~/.talos/mcp-config.enc \
    --talosconfig-identity ~/.ssh/id_ed25519

Set the talosconfig path with --talosconfig or TALOSCONFIG. The default is ~/.talos/config.

See docs/config.md for every talosconfig field, the discovery block, credential roles, the unlock options for encrypted secrets, the config import, encrypt, decrypt, and check commands, deployment recipes, and troubleshooting.

Configure an MCP client

Local stdio server

For clients that use a JSON MCP configuration, add an entry like this:

{
  "mcpServers": {
    "talos": {
      "command": "talos-mcp",
      "args": ["mcp"],
      "env": {
        "TALOSCONFIG": "/absolute/path/to/talosconfig"
      }
    }
  }
}

If the talosconfig has encrypted fields, also set an unlock source, for example "TALOSCONFIG_IDENTITY": "/absolute/path/to/.ssh/id_ed25519". The mcp command never prompts for a passphrase.

Remote HTTP server

Start the streamable HTTP server with:

talos-mcp server \
  --talosconfig /absolute/path/to/talosconfig \
  --listen-address :8080

The MCP endpoint is http://host:8080/mcp, and a health check is served on /healthz.

For a remote client, use an HTTPS URL that ends with /mcp:

{
  "mcpServers": {
    "talos": {
      "type": "http",
      "url": "https://talos-mcp.example.com/mcp"
    }
  }
}

The server does not provide user authentication for the HTTP endpoint. Every caller uses the credentials from the server's talosconfig, so protect the endpoint with a network policy or an authenticating proxy.

A Helm chart for Kubernetes is available in charts/talos-mcp.

Restart or reload the MCP client after you save its configuration.

Running

Common flags

The mcp, server, and tools commands use the following flags. Each flag can also be set with an environment variable. A command-line flag has higher priority than an environment variable.

Flag

Environment variable

Default

Description

--talosconfig <path>

TALOSCONFIG

~/.talos/config

Path to the talosconfig file.

--context <name>

TALOS_CONTEXT

All contexts

Use only this one talosconfig context.

--talosconfig-identity <path>

TALOSCONFIG_IDENTITY

None

age identity or OpenSSH private key that decrypts encrypted fields. Repeatable.

--talosconfig-passphrase-file <path>

TALOSCONFIG_PASSPHRASE_FILE

None

File with the passphrase of encrypted fields.

--talosconfig-askpass <program>

TALOSCONFIG_ASKPASS

None

Program that prints the passphrase of encrypted fields.

--extensions <list>

EXTENSIONS

all

Comma-separated tool groups to enable: cluster, node, or all.

--allow-destructive

ALLOW_DESTRUCTIVE

false

Allow destructive tools, such as talos_node_reboot.

--log-level <level>

LOG_LEVEL

info

Log level: debug, info, warn, or error.

--log-format <format>

LOG_FORMAT

text

Log output format: text or json.

The server command also accepts --listen-address (LISTEN_ADDRESS, default 127.0.0.1:8080) and --require-all-contexts (REQUIRE_ALL_CONTEXTS). The container image sets LISTEN_ADDRESS=:8080. The tools command accepts --output (-o) with text, json, or yaml. Its default is text.

For example, run the stdio server with JSON logs:

talos-mcp mcp \
  --talosconfig /absolute/path/to/talosconfig \
  --log-format json

Test the configuration

Check the talosconfig and its encrypted fields:

talos-mcp config check --talosconfig /absolute/path/to/talosconfig

List all available tools:

export TALOSCONFIG="/absolute/path/to/talosconfig"
talos-mcp tools

Call a tool directly:

talos-mcp tools talos_clusters_list
talos-mcp tools talos_clusters_describe cluster=prod-eu
talos-mcp tools -o json talos_node_logs cluster=prod-eu node=10.0.0.10 service=kubelet tail=50

License

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.


Talos Linux is a trademark of Sidero Labs, Inc.

Related MCP Connectors

Related MCP Servers

  • A
    license
    A
    quality
    C
    maintenance
    Enables AI assistants to interact with and manage Kubernetes clusters, supporting operations on pods, deployments, services, configmaps, secrets, namespaces, metrics, and events with built-in safety features for destructive actions.
    9
    7 npm
    1
    MIT
  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables AI agents to diagnose and recover Kubernetes clusters when the API server is unreachable, using SSH access to nodes and tools like etcdctl, systemctl, and kubeadm with 63 read-only and destructive tools.
    MIT
  • A
    license
    A
    quality
    B
    maintenance
    Gives AI assistants read-only access to a Kubernetes cluster's observability stack — Loki, Prometheus, Tempo, the Kubernetes API — plus AWS telemetry from CloudTrail, Config, ECR, and RDS. 23 tools for querying logs, metrics, traces, Kubernetes events, deployments, and audit history in plain language.
    23
    Apache 2.0