Skip to content

Sanity Context

Content is Context

We make it easy to give agents the right context, so they get it right

Teams powering context with Sanity

  • Braze
  • Anthropic
  • Brex
  • Shopify
  • Linear
  • Replit
  • Expedia Group

The problemThe problem

Agents deserve better contextAgents deserve better context

Dark-themed user interface displaying an AI assistant answering data encryption queries within a chat thread next to a documentation knowledge base.

When your content is all over the place, agents become inaccurate, slow, and expensive. Context provides an optimized source for agents, with human control.

The BenefitsThe Benefits

Your content, your rulesYour content, your rules

  • The right answers, faster

    Context organizes your sources before your agent ever asks.

  • You run the show

    The people who own your knowledge, voice, and tone manage what agents know and how they sound.

  • Knowledge governance

    Every fact cites its source, and insights show what agents were asked, how well they answered, and which entries they used.

How it worksHow it works

From scattered to grounded context, in minutesFrom scattered to grounded context, in minutes

A stale source, cleaning_2019.pdf, is connected, then caught in review, cut off and dropped: the old PDF that says must hand wash our products. Two sources, both checked as accurate, feed one knowledge base entry, cleaning care. From the faqs: the water, never use warm water. From the video tutorial: the pro tip, dry with a dry towel on low heat in the drier. The entry also lists care instructions, drier or hang drying is fine. It cites both sources. Over MCP, the agent reads the corrected context instead of wasting time on wrong content.
cleaning care

# How to Clean our Products

care instructionsdrier or hang drying is fine
droppedold PDF that says must hand wash our products
Sources
MCPreads the corrected context instead of wasting time on wrong content
  1. 1

    Ingest

    Websites you own, uploaded files, or Sanity datasets

  2. 2

    Organize

    Your messy content is organized for agents. Your team resolves any contradictions found.

  3. 3

    Connect

    One MCP, any model

Quick startQuick start

Run this prompt, get an agentRun this prompt, get an agent

  1. Run the prompt

    Ask your coding assistant to build the agent.

  2. Get a working agent

    Your coding assistant asks a few questions, then builds an agent for your project.

Your coding assistant asks a few questions, then builds an agent for your project in minutes.

Already have an agent? Just add your Context MCP. See the setup docs.

Use casesUse cases

Power every type of agentPower every type of agent

Dark-mode software dashboard showing connected knowledge bases beside an active support agent chat window providing code-based troubleshooting.

For builders experimenting and serious teams

  • External agents

    Ecommerce assistants, support agents, web chatbots

  • Internal agents

    Internal wiki agents, training bots, onboarding agents

  • Automation loops

    Agentic applications running workflows or processes

FeaturesFeatures

What’s in the boxWhat’s in the box

  • Tools

    Context gives your agent tools to query your content directly or read a Knowledge Base built from it.

  • Insights

    Optionally log and analyze every conversation with a success score, sentiment, and gaps.

  • Conflicts

    When two facts disagree, you choose the right one before a customer asks.

  • Sources

    Crawl websites, upload files, connect your Sanity data, or send markdown using the CLI.

  • Scope access

    Decide which content each MCP has read access to.

  • Any model, any harness

    Context provides your content via an MCP. You bring the LLM and agent.

  • Editorial control

    Use documents to control parts of the system prompt, such as voice and tone, so editors can tweak things without developers.

  • Embeddings

    Search content for semantic similarity with built-in embeddings.

  • Multiple MCPs

    Create different MCPs for different use cases.

  • CLI and client

    Manage it from code and build automations.

Pricing

Start for free

  • Free$0

    Includes

    • 1 Knowledge Base
    • 10,000 source chunks

    All existing Sanity plans include 1 free Knowledge Base

    Get started free
  • Self-serve$299/mo

    Includes

    • Up to 5 Knowledge Bases
    • 50,000 source chunks
    • Weekly or monthly scheduled refresh
    Get started
  • Enterprise Custom

    Includes

    • 5+ Knowledge Bases
    • 100,000 source chunks
    • Weekly or monthly scheduled refresh
    Speak to sales

FAQs

Do I need to use Sanity's CMS or move my website?

No. Context works on its own. Bring your content and keep your stack.

What models and agents does this work with?

Any that speak MCP. You bring the agent and model and Sanity provides you with an MCP. So Claude, OpenAI, Gemini, or any other model works. You can also use any harness such as Mastra, eve, LangChain, and more.

Which mode should I use? GROQ or Knowledge Bases?

Query live content (GROQ) when you need exact filters or answers that change by the minute, like price, size, and stock. Use a Knowledge Base for policies, manuals, and help content, which a build can reconcile and cite ahead of time. Many agents use both.

What content can I bring?

Sanity Datasets, websites, files, and plain text. You can add these via the UI or CLI for automation.

Can agents change my content?

No. Context is read-only.

What does it cost?

It’s free, including creating a knowledge base. Upgrade to create more Knowledge Bases, refresh them automatically, and more. See pricing.

How is this different from a vector database?

Vectors enable you to find similar content, Sanity Context enables you to find the exact content that matches the input. Through GROQ mode your agent can use exact filters (price < $25) and optionally rank the results by semantic similarity (“durable goggles”) using Dataset Embeddings in a single query and without managing an embeddings pipeline. With Knowledge Base mode, your agent finds precise facts in your corpus (“return policy is 30 days”).

Sanity MCP Server or Context MCP?

Use the Sanity MCP Server when an agent needs to create, edit, or manage content, like an AI assistant that works on your project. Use Context MCP when an agent answers questions from your content, like a shopping or support agent. Context MCP is read-only and scoped to the content you choose.

How do I personalize per user?

Pass user context in the system prompt, and a bearer token for per-user calls.

What is a source chunk?

A source chunk is a unit of source content that is ingested during the Knowledge Base build. One source chunk is approximately 250 words or 1,500 characters.

How do I estimate source chunks?

A source chunk is how we measure the amount of content a Knowledge Base reads. To get an estimate, divide your word count by 250.

SourceExampleApproximate source chunks
WebsiteShort page ~300 words1-2
WebsiteArticle ~1k words4
WebsiteLong guide, ~2k8
Sanity datasetProduct entry, ~150 words1
Sanity datasetLong article, ~2k words8
Uploaded filePDF or document 2 per page

How many Knowledge Bases can I have?

Every Sanity plan includes one Knowledge Base on the Free plan. An organization admin can upgrade to the Self-serve plan that includes five Knowledge Bases with a higher source chunk limit. For teams that need more than five or a higher ingest limit, contact Sales for Enterprise options.


Turn your content into context

Ship a smart agent that answers from your content