PLUGIN / SANITY-IO

Sanity Agent Toolkit

How to build on Sanity, written for the assistant doing the typing: the content model, the queries, the front end, and the move in from an older system.

OFFICIAL★ 161 GitHub starsMITRead the official docs ↗

Labsco Summary

ShowHideGuides an assistant through building on Sanity, from content models and queries to front ends and migrations, beside the live connection Sanity prefers.

The connection comes first, the advice second.

Sanity's readme does not open with the written guidance. It opens with two different offers, and recommends the first:

a connection to your real Sanity project, which reads and changes content, datasets, schemas and releases
8 pieces of written guidance an assistant reads before it touches anything

The second is positioned for assistants that cannot hold a connection, or as something to run beside one.

Eight pieces, by stage of the work.

The 8 divide by where the work has got to:

2 on standing the project up: Sanity's own house practices, and how to shape content so it survives being reused elsewhere
3 on getting content in and back out: in from another system, out to a browser
3 on what happens after it is live: being found by search engines and by assistants answering out loud, testing one wording against another, and upgrading the editing Studio

It assumes you are on Sanity.

None of it is content-system-neutral:

all 8 are written around Sanity's own query language, its text format and its editing setup
even the search guidance is framed around a Sanity-backed front end rather than content systems at large
the migration one covers a real project end to end — extraction, conversion, redirects, validation, cutover

On another content system with no plan to move, none of it applies as written.

Skills

Decisions taken before much content exists: how the content types are shaped, what becomes a reference and what stays nested, and the conventions for queries, generated types and the editing experience that everything built later inherits.

content-modeling-best-practicesHow to organise the text and images in a content system so they can be reused on a new page, app or channel later, rather than being shaped to fit the one screen they were written for.sanity-best-practicesSanity's own engineers wrote down how they want a Sanity project built, and this is that advice — the content structure, the queries, the frontend wiring and the editing experience.

Commands

/sanity:deploy-schema

Sends a changed content model up to the project, with the check that it actually landed built into the same step.

/sanity:sanity-review

Turns the assistant on Sanity code that already exists and has it report where the code departs from the practices in these skills: the query, the content types, the visual editing wiring.

/sanity:sanity

Asks the assistant to lay out what it can now do with Sanity and which help topics it can open — the way in when you do not yet know what to ask for.

/sanity:typegen

Hands the whole type-generation run to the assistant, including the part afterwards where something has gone wrong and has to be read and put right.

Reads your real content, and can publish it.

CONNECTS TOYour Sanity project — its content, schemas and releases

This is the route the readme puts first: the assistant stops working from a description of your content model and starts reading the project itself.

Configured by hand, it authorises through a sign-in; a bearer token can be set instead. Sanity's own command-line tool will also set the whole thing up from whichever account that tool is already signed in as.

What it reads.

documents, by query

the schema that is deployed, and the Studio apps a project has

content releases, and which user the connection is acting as

Sanity's documentation, and the agent rules kept in this repository

What it changes.

creates drafts, and edits documents that already exist

publishes a document, and takes it back off again

deploys a schema, and a Studio bound to that schema

opens a release, and writes versions of documents into it

produces images, and alters ones already there

Signing in is a decision of its own. The written guidance does its job with this connection never set up.

Three of those draw on Sanity AI credits rather than just time: producing an image, altering one, and asking for a version of a document to be written for you.

Every tool the server offers, and what each one changes:Sanity MCP server

https://mcp.sanity.io

INSTALL

Either route puts the written guidance in; the plugin is the one that brings the project connection with it:
terminal, or plugin.

TERMINAL

Paste one line in a terminal

npx skills add sanity-io/agent-toolkit

PLUGIN MARKETPLACE

Add the marketplace, then install the plugin

Typed inside the agent's own prompt, not in a terminal. The marketplace is called sanity-agent-toolkit, which is the part after the @.

01/plugin marketplace add sanity-io/agent-toolkit
02/plugin install sanity@sanity-agent-toolkit