Skip to main content

Connecting your AI assistant to Instill

Learn how to connect any MCP-compatible AI assistant to your Instill workspace. Popular tools include Claude, Cursor, VS Code, Windsurf, and Zed — but any client that supports the Model Context Protocol will work.

M
Written by Moto Mo
Updated this week

Who is this for?

MCP is useful for anyone who works with an AI assistant outside of the Instill:

  • Developers — manage collections, query data, or update schemas from Cursor, VS Code, or Claude Code without switching windows

  • Analysts — pull data from Instill into AI tools for deeper investigation

  • Team leads — use AI assistants to quickly check project status or search across collections

You don't need to write code to use MCP — the AI assistant handles everything through natural language.

What can you do once connected?

Once your AI assistant is connected, it can:

  • Browse all your organizations, projects, and collections

  • Search across your documents and files

  • Create, update, and delete projects, collections, rows, and columns

  • Run analytics queries (sort, filter, group by) on your data

  • Manage row and column permissions

All write operations require your confirmation before they execute — nothing changes without your approval.

Step 1: Get your API token

If you plan to use the API Token method (instead of OAuth), you'll need a token first:

  1. Go to Settings > API Tokens in the Instill AI app

  2. Click Create Token

  3. Give it a name and click Create

  4. Copy the token — you'll need it for setup

Step 2: Add the MCP server to your AI tool

The Instill AI MCP endpoint is:

There are two ways to connect: OAuth (recommended) and API Token (manual).

1. Connect with OAuth (recommended)

OAuth is the simplest option. Your AI tool opens a browser for you to log in — no token needed.

Windsurf — add to ~/.windsurf/mcp.json:

{
"mcpServers": {
"instill-ai": {
"serverUrl": "https://mcp.instill-ai.com/mcp"
}
}
}

Zedsettings.json:

{
"context_servers": {
"instill-ai": {
"transport": "http",
"url": "https://mcp.instill-ai.com/mcp"
}
}
}

2. Connect with API Token

If your AI tool doesn't support OAuth, use an API token instead.

Claude Code

claude mcp add instill-ai \
--transport http \
--url https://mcp.instill-ai.com/mcp \
--header "Authorization: Bearer YOUR_API_TOKEN"

Cursor

{
"mcpServers": {
"instill-ai": {
"url": "https://mcp.instill-ai.com/mcp",
"headers": {
"Authorization": "Bearer YOUR_API_TOKEN"
}
}
}
}

For VS Code, Windsurf, and Zed — use the same pattern: add "headers": { "Authorization": "Bearer YOUR_API_TOKEN" } to the server config.

Reducing token usage with scopes

Instill AI provides around 49 MCP tools. Loading all of them at once uses a lot of context tokens. You can use the scope parameter to load only what you need:

Scope

Tools

Best for

core

10

Browsing projects, collections, and files (read-only)

core,data

36

General workspace operations (read + write)

core,insight

17

Analytics and data exploration

core,data,insight

43

Full workspace + analytics

To use a scope, append it to the URL:

Working with multiple organizations

If you belong to more than one organization, your AI tool will ask you to choose which one to use. Simply specify the organization ID when prompted.

If you belong to only one organization, it's selected automatically — no extra step needed.

In Instill, all resources (projects, collections, files) belong to an organization workspace. You cannot create resources under a personal workspace — choose an organization workspace when prompted.

Notes

Instill is currently in private launch — access is available by invitation only, with a soft launch to follow. To request access or learn more, contact [email protected] or visit instill-ai.com.

Did this answer your question?