Hygraph supports connecting AI assistants directly to your project as connectors, so you can ask them to read, draft, and publish content without a developer setting anything up for you. This uses the same Hygraph MCP server that developers configure with a Permanent Auth Token for tools like Cursor or VS Code. You're using the same underlying connection, just through your own Hygraph login instead of a token.

#Connect to Claude

1. Go to the Hygraph MCP connector in the Claude directory. Otherwise, if you're not redirected, in Claude, go to Settings > Connectors, find Hygraph MCP, and select Connect.
2. Sign in to Hygraph when prompted, and click Accept.
   • You might see a message that you Allow offline access. This enables Claude to keep working on Hygraph tasks in the background or on a schedule, without the app open.

Once connected, ask Claude what Hygraph projects it can see. It only shows the ones your Hygraph account already has access to.

#What a connected AI assistant can do

A connected assistant's access follows your own Hygraph user permissions, the same permissions that decide what you can see and do in Studio. Connecting an assistant doesn't grant it anything beyond that. If you can't publish a model in Studio, the AI assistant can't publish it for you either.

Because one login can reach every project you're a member of, ask the assistant to list your projects and pick an environment before you give it a task.

A connected assistant can read your project's AI Guidelines before it drafts or changes content, the same guidelines that shape output from AI Agents and AI Assist. See AI Guidelines for what each category covers.

It can also trigger any agent in your project that's configured for Use in Agent runs, the same setting that makes an agent available from the content table and entry form. Agent runs triggered this way land in DRAFT and wait for your review, exactly like a manual run in Studio. See Trigger agents for how to review the output either way.

#Tool permissions

Most AI assistants ask before they use certain tools against your Hygraph project. In Claude, you choose Allow once to approve a single action, or Always allow to skip that prompt for the rest of the conversation.

Be careful about the permissions you grant to the execute_graphql tool. It can read anything you can read and make any non-destructive change you're permitted to make, including bulk updates across many entries at once.

#What a connected assistant can never do

No matter which assistant you connect or which tools you approve, it cannot delete content or schema. If you want something deleted, do it yourself in Studio.

Everything a connected assistant creates or changes lands in DRAFT first. Nothing reaches your published content until an entry is published, either by you or by a publish action you approved.

#Troubleshooting

#The assistant asks which project or environment to use every time

That's expected. One login can reach every project you have access to, so the assistant asks you to confirm which one before it acts. Answer once and it keeps using that project and environment for the rest of the conversation.

#Hygraph sign-in succeeds but the assistant says it can't do something

Confirm you can do that action yourself in Studio. The assistant's access is your access. If you should be able to do it and can't, check with your admin.

#I don't see Hygraph in Claude's connector directory

Ask a Team or Enterprise Owner to add the Hygraph connector under Organization settings before you try connecting.

#What's next

• AI tools for editors: Overview of AI Agents and AI Assist, the tools built into the content editor.
• Trigger agents: Trigger agents from the Content Table or Content Form, and review their output.
• Use AI Assist: Run AI Assist tasks on individual fields within an entry.
• AI Guidelines: Understand what guidelines are and how to request changes from your admin.