Whiskers
Whiskers — Zero-Trust MCP Control Plane for Docker & Kubernetes. Tactile observability, mTLS mesh, blast-radius guardrails, and human-in-the-loop approvals for AI agents.
Install / Use
claude mcp add LupusMalusDeviant -- npx -y github:LupusMalusDeviant/WhiskersIf the server publishes to npm under a different name, use that package instead — check the repo README.
MCP Server
Model Context Protocol server
Quality Score
Category
AI & Machine LearningSupported Platforms
Our assessment of Whiskers
Whiskers scores 83/100 on our quality scale, 656th of 948 AI & Machine Learning skills we index.
Its MCP Server is 14 KB long, well organised into 13 sections with 7 code examples: a thorough specification that gives an agent plenty to work with.
It has 3 GitHub stars, so there is little community track record yet; judge it on its content.
Maintenance, license and trust
- The repository was last updated today, so Whiskers is actively maintained.
- It is released under the Apache-2.0 license, a permissive license that allows use, modification and commercial use with attribution.
- Its trust signals score 92/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Whiskers compared with similar skills
All 4 of these similar skills score higher than Whiskers; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| Whiskers (this skill)by LupusMalusDeviant | 83 | 3 | today | MCP Server |
| claude-memby thedotmack | 100 | 114.6k | today | CLAUDE.md |
| Agent-Reachby Panniantong | 100 | 110.0k | today | CLAUDE.md |
| Understand-Anythingby Egonex-AI | 100 | 91.3k | 1d ago | CLAUDE.md |
| headroomby headroomlabs-ai | 100 | 79.1k | today | CLAUDE.md |
Frequently asked questions
- How do I install Whiskers?
- Run
claude mcp add LupusMalusDeviant -- npx -y github:LupusMalusDeviant/Whiskers. The install tabs above show the steps for each supported agent. - Which AI agents does Whiskers work with?
- It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
- Is Whiskers safe to use?
- It is Apache-2.0-licensed and scores 92/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is Whiskers still maintained?
- The repository was last updated today, so Whiskers is actively maintained.
Skill content
View source on GitHub
Whiskers
Self-hosted MCP control plane for Docker hosts and Kubernetes clusters. Humans and AI agents call scoped tools instead of holding SSH keys or the Docker socket. Every call is authenticated, scope-checked, guardrail-evaluated and logged.
Status: beta. Not API-stable before 1.0. Run it on a trusted network and read SECURITY.md.
Scope of this README: main, i.e. the first release after 0.15.0. Releases up to 0.15.0 differ (installer defaults
to latest, no generated vault key, no VAULT_KEY_FILE, no setup token); for those use the
README at v0.15.0.
Facts
| | |
|---|---|
| Interfaces | Web UI (Blazor Server) and MCP over Streamable HTTP at /mcp. No hosted or cloud endpoint. |
| Sign-in | Local accounts, Google, generic OIDC. AUTH_DISABLED=true (container variable Auth__Disabled) makes every request an Admin (trusted LAN trials only). |
| MCP keys | Read, Write or Admin scope, or an explicit tool list; checked on every call. Stored as SHA-256 hashes, shown once. |
| Guardrails | Per tool Allow / Confirm / Block, evaluated in code at the tool boundary. Confirm runs only the exact call a human approved. |
| Records | Agent History (every tool call) and a hash-chained Audit Log. Checkpoints are signed, and the trail is tamper-evident, only when a vault key (VAULT_KEY or VAULT_KEY_FILE) is set. |
| Host access | Onboarding uses SSH once. Onboarded hosts then run over a WireGuard mesh (Tailscale/NetBird), the Docker API over mTLS behind a verb-whitelisting socket proxy, and host commands via a one-shot nsenter container, with no standing SSH key. A server added as a plain SSH connection and never onboarded keeps using SSH (its key in ssh-keys/) until it is migrated. |
| Targets | Docker hosts, k3s/Kubernetes clusters (kubeconfig), Hetzner and Hostinger cloud APIs. |
| Storage | /app/data (WHISKERS_DATA_DIR). SQLite by default, PostgreSQL optional. Secrets vault AES-256-GCM with VAULT_KEY / VAULT_KEY_FILE; SSH and mTLS client keys are protected by file permissions only. |
| Images | ghcr.io/lupusmalusdeviant/whiskers, amd64 + arm64, Trivy-gated, SBOM + provenance, keyless cosign signature. |
| Stack | C# / .NET 10, ASP.NET Core, MudBlazor, EF Core, SignalR. License Apache 2.0. |
The full tool list with permission levels is generated from the code: docs/mcp-tool-catalog.md.
Install
VER=X.Y.Z # a release from https://github.com/LupusMalusDeviant/Whiskers/releases
curl -fsSLO "https://github.com/LupusMalusDeviant/Whiskers/releases/download/v${VER}/install.sh"
curl -fsSLO "https://github.com/LupusMalusDeviant/Whiskers/releases/download/v${VER}/checksums.txt"
sha256sum --check --ignore-missing checksums.txt && bash install.sh
The installer installs exactly that release, generates the vault key on a new install (VAULT_KEY_FILE), verifies the
image signature if cosign is installed, writes ./whiskers/ and waits until healthy. The UI is at
http://127.0.0.1:5100. Flags: install.sh --help.
Without the installer:
VER=X.Y.Z
docker run -d --name whiskers -p 127.0.0.1:5100:8080 \
-v whiskers-data:/app/data -v /var/run/docker.sock:/var/run/docker.sock \
"ghcr.io/lupusmalusdeviant/whiskers:${VER}"
This has no vault key; set VAULT_KEY_FILE before you store secrets. Compose, Helm, reverse proxy, Tailscale and the
hardened profile: docs/installation.md.
First sign-in
docker exec whiskers cat /app/data/setup-token.txt # installer or docker run
docker exec serverwatch cat /app/data/setup-token.txt # release docker-compose.yml (container keeps its pre-rebrand name)
Enter the token in the setup wizard; it creates the first Admin. Unattended alternative: WHISKERS_ADMIN_EMAIL +
WHISKERS_ADMIN_PASSWORD_FILE. Local accounts have no MFA in 1.0; for production, sign in through your OIDC provider
with MFA.
Connect an MCP client
Create a key under Settings → MCP, then:
export WHISKERS_MCP_API_KEY=…
claude mcp add --transport http --scope project whiskers http://localhost:5100/mcp --header 'Authorization: Bearer ${WHISKERS_MCP_API_KEY}'
Any other client:
{
"mcpServers": {
"whiskers": {
"type": "http",
"url": "http://localhost:5100/mcp",
"headers": { "Authorization": "Bearer <WHISKERS_MCP_API_KEY>" }
}
}
}
Crush and more: docs/installation.md → MCP clients. Scopes, prompts and resources: docs/mcp.md.
Verifying the image
VER=X.Y.Z
cosign verify "ghcr.io/lupusmalusdeviant/whiskers:${VER}" \
--certificate-identity-regexp '^https://github.com/LupusMalusDeviant/Whiskers/.github/workflows/release.yml@refs/tags/v' \
--certificate-oidc-issuer https://token.actions.githubusercontent.com
Images before v0.12.1 are not cosign-signed. Supply-chain details: docs/security-architecture.md.
Configuration
Everything is optional. Where a setting goes depends on the install path:
- Release
docker-compose.yml: put the.envnames below in a.envbeside it; the compose file maps them to the container variables. - Installer: on a new install its compose file sets
TZ,VAULT_KEY_FILEand, with--data,WHISKERS_DATA_HOST_DIR. Add container variables asenvironment:entries in./whiskers/docker-compose.yml. docker run: nothing is set; pass container variables with-e.
| .env name (release compose) | Container variable | Effect |
|---|---|---|
| VAULT_KEY_FILE / VAULT_KEY | same | Encrypts stored secrets. Losing it makes them unreadable. The installer sets VAULT_KEY_FILE on a new install. |
| — (release or repository compose file: deploy/docker-compose.postgres.yml overlay from the repository; installer: environment: entries) | WHISKERS_DB_PROVIDER, WHISKERS_DB_CONNECTION (or _FILE) | sqlite (default) or postgres. The overlay needs WHISKERS_DB_PASSWORD in .env. |
| OIDC_ENABLED, OIDC_AUTHORITY, OIDC_CLIENT_ID, OIDC_CLIENT_SECRET | Oidc__Enabled, Oidc__Authority, Oidc__ClientId, Oidc__ClientSecret | Sign-in through your identity provider. |
| HOST_BIND, HOST_PORT | — (port mapping; installer: --bind, --port) | Published address; default 127.0.0.1:5100. |
| PATH_BASE | PathBase | Subpath behind a reverse proxy. |
| — (not mapped; add an environment: entry) | ForwardedHeaders__TrustedNetworks__0 | Proxy whose X-Forwarded-* headers are believed; loopback only by default. |
All keys: docs/configuration.md and .env.example.
Accepted limits (1.0)
Deliberate limits of Whiskers 1.0 (product contract §5):
- Root-equivalent administration. An Admin, an Admin-level key or a compromised instance is root on every managed host. Whiskers governs who reaches that power and records it; it does not make it smaller. A shell on the Whiskers host itself is outside the contract.
- Single instance. One process and one data directory; no replica, no failover. Recovery comes from self-backups kept off the host.
- Global roles. Viewer, Operator and Admin (and key levels) apply to every managed server; there is no per-server or tenant scoping.
- No terminal transcript. Opening a terminal is authorized, kill-switch-checked and audited; what is typed inside is neither policy-checked nor recorded.
- Out-of-band changes are invisible. Changes made directly on a host, in a cloud console or by another tool are not governed or attributed.
- No second factor for local accounts. Production sign-in goes through your identity provider with MFA.
Compose exposure checks: remaining risks
Compose deployments are decided on their exposures (published ports, host network, privileges, route labels; contract
0006, INV-62 — details in docs/features.md). Not caught,
or caught only after up:
- Traefik
exposedByDefault: a Traefik that routes every container without a label publishes services this check never sees. - Tailscale sidecar
serve/Funnel configured inside a sidecar's own state orserve.json, and cloudflared tunnels, publish without a label on the deployed service. - The re-tag window: an image re-tagged by someone else between the image inspection and
up --pull neveris caught only by the re-check afterup. - The tag race: putting a tag back after a refused run checks that it still points where this run left it; a tag moved by another run in between is left alone (and so not restored).
container.updateandautopatch.runrecreate containers from new images without this check.- Late daemon work after a timeout: a pull, build or
upthe daemon finishes after Whiskers stopped waiting is not re-checked. - The 5-minute budget of a compose run (per command) can be too short for large builds; Git deploy keeps its own 300 s / 600 s / 300 s.
- Out of scope: the Deploy page's compose mode (
ComposeFileParser) andDeploymentRequest/deploy_appare not decided on their exposures.
Upgrade note (contract 0006). The new policy defaults change every policy's version hash, so open approvals and
the policy snapshots of pinned runs expire at the upgrade. docker compose pull --ignore-buildable needs Docker
Compose v2.15 or newer on every managed host. A label key, a port, a network or a privilege field from an .env
variable that is not set now refuses the plan. Redeploys of projects with build: services rebuild with --pull, and
Git deploys now pull images as well.
Operating it
Production checklist · Upgrade and rollback · Backup and recovery · Troubleshooting · all operations guides
Development
Requires the .NET SDK version in global.json.
dotnet build Whiskers.slnx -c Release
dotnet test Whiskers.slnx -c Release
dotnet test src/Whiskers.Tests/Whiskers.Tests.csproj --filter Category=RequiresDocker # PostgreSQL suites: need Docker, skipped locally without it, mandatory on CI
dotnet run --project src/Whiskers/Whiskers.csproj # http://localhost:5041 (launchSettings.j
Truncated for display — read the full file on GitHub.
Related Skills
claude-mem
114.6kPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
Agent-Reach
110.0kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
Understand-Anything
91.3kGraphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.
headroom
79.1kCompress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
