Viglet Shio: the open source, agent-ready CMS
Viglet Shio is an open source, agent-ready CMS (content management system).
A person can run all of it from the console, or hand it to an AI coding agent such as
Claude Code, which builds and maintains the site
through MCP tools, content as files and blueprints; the curator approves in
a React console with a Visual Editor and preview links; and a Content
Delivery API (REST + GraphQL, with TypeScript SDKs) delivers the content to
Next.js, Astro, React, Vue, Svelte or any front end you like.
Shio hosts no model and no prompts. It is the substrate an external agent
drives, so the differentiator is protocol quality (discoverability, determinism,
verifiability, token economy), not intelligence.
🌐 Product site: shio.viglet.org ·
📚 Documentation: shio.viglet.org/docs ·
🏛 Viglet open source: viglet.org ·
🔎 Search & personalization: turing.viglet.org
Shio (pronounced [strong /ʃiː/ · weak /ʃɪ/ o]).
Two ways to start
Agents draft, a person approves.
-
Use it hosted, free: create a free account on viglet.cloud.
Nothing to install.
-
Run it on your server:
docker run -p 2710:2710 -e SHIO_ADMIN_PASSWORD=change-me ghcr.io/openviglet/shio:latest
Then open http://localhost:2710 and sign in as admin. The Quick start has the rest.
Table of contents
What is Shio?
Shio is a Java + Spring Boot CMS with a React administration
console and a TypeScript delivery layer. It models content the way a CMS
always has (sites, folders, posts, post-types, widgets) and then exposes every
one of those capabilities on a surface an agent can actually drive:
| Surface | What it is |
|---|
MCP server (POST /mcp) | 12 Model Context Protocol tools: shio_context, shio_find, shio_read, shio_write, shio_apply, shio_publish, shio_assets, shio_changes, shio_verify, shio_digest, shio_remember, shio_marketplace |
Agent gateway (/api/v2/agent/**) | A manifest, a context pack, planned batches, teaching errors, a review queue, a diagnostics endpoint, and which fields a curator could edit |
Content as files (shio/content/**) | The whole site projected to front-mattered Markdown on disk, with a three-way merge back, so Edit, Write and Grep are all an agent needs to author |
CLI (@viglet/shio) | 48 verbs: pull, push, apply, publish, find, verify, preview, clone, convert, types, dev, deploy, patterns, status, index, redirects, export, … |
Content Delivery API (/api/v2/cda/**) | Stable REST + GraphQL reads, token-scoped, with zero-dependency TypeScript SDKs, plus two narrow writes: the Visual Editor's save-back, and form submissions landing as drafts |
| React console | The curator's half: content browser, 15 form widgets, Visual Editor, review queue, preview links |
Keywords, so you can find this again: agent-ready CMS, headless CMS, agent-native CMS, AI CMS,
MCP server, Model Context Protocol, Claude Code, content management system, Java
CMS, Spring Boot CMS, Next.js CMS, React CMS, GraphQL CMS, content as code,
GitOps CMS, open source CMS, self-hosted CMS.
Why an agent-native CMS?
Every mainstream headless CMS was designed for a person clicking through a web
UI, and then had an API bolted on. When an agent drives one of those, it burns a
whole session discovering what exists, guesses at IDs, gets a 400 Bad Request
with no instruction in it, and cannot prove its work without asking someone to look at
a screen.
Shio inverts the order. Every capability lands MCP → files/CLI → REST →
console: a console-only feature is considered incomplete. Discovery is one
call. Errors carry fix, allowed, didYouMean and example. Response sizes
are a measured budget asserted in CI, so a token regression fails the build
like any other test. And the curator is never surprised: writes are draft by
default, publishing is explicit, and every agent change lands in a review queue
with attribution.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
