Engine-agnostic, agent-ergonomic Docker MCP server in Rust — token-bounded reads, gated destructive writes, deterministic diagnostics
View on GitHub ↗An engine-agnostic, agent-ergonomic Docker MCP server in Rust.
Kagoni — from 籠 (kago), Japanese for basket, cage or crate. 籠に means "in the basket": a bounded space you pack carefully because you cannot make it bigger. That is the whole design.
Kagoni exposes container lifecycle, logs, stats and Compose as MCP tools — but every read is bounded and summarizing by default, every destructive write is gated, and it talks the plain Docker Engine socket, so it drives Docker, OrbStack, Podman or Colima through the same code path — see what has actually been tested.
It is not a monitoring daemon. It is a control surface for a human-in-the-loop agent.
This space is not empty — thin CRUD MCP wrappers around docker ps are everywhere, and autonomous remediation daemons already exist. Kagoni is narrower than either:
Kagoni Thin CRUD wrappers Autonomous daemons ------------ Interaction human-in-the-loop via MCP human-in-the-loop autonomous loop Read discipline bounded by design relays raw daemon output n/a Diagnosis deterministic, evidence-carrying none LLM-in-the-loop Engines any Docker-API engine usually Docker only usually Docker only
The one idea worth the code is token-budget as the primary design constraint. A crash-looping container emits the same stacktrace five hundred times; relaying that verbatim spends a context window to say one thing. Kagoni's job is to consume the firehose and hand back a digest.
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/nhatvu148--kagoni.html)
A Datalog engine for LLM agent memory: stratified rules, provenance-tracked facts, incremental derivation, and an MCP server that lets your harness use it as a shared brain.
mrpulor-gh/nuphus-mcpDesktop automation MCP server — computer use for any AI agent: control screen, windows, mouse/keyboard, and Chrome via Model Context Protocol (stdio)
KayanoLiam/ContextXA free remote MCP server offering normal (Grok 4.3) and deep (Grok 4.20 Multi-Agent) search. No API keys required, streaming support prevents gateway timeouts.
omaekumiko2-create/kruLocal-first MCP password and credential manager for AI agents. Use passwords, API keys, SSH identities, and TOTP without exposing hidden plaintext to the model.
pinoox/neuromeshThe Biomimetic Context Engine & Neural Runtime for AI Coding Assistants
naw103/foremergeCatch intent conflicts before code conflicts. The open-source coordination protocol for coding agents, built above Git.
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.