AnythinkDocs
Start here
What you can build
Guides
Ship and operate
Reference
What's new
See something missing?
powered by

Chat with your backend data using AI

In about ten minutes, ask the Anythink AI chat about your records, have it change one, then connect Claude over MCP to do the same from your editor.

Last updated

You'll ask an AI model about your project's data, have it change a record, and then connect Claude over MCP so you can do the same from your editor. The AI chat and the MCP server can both act on your data, not just answer questions, and both do it with your permissions. For how the pieces fit together, see How AI chat and MCP work.

Before you start

You need a project with an orders entity that has a status (text) and a total (decimal) field, and a few records in it. If you haven't created entities before, see Model your data.

To set up the same data from a terminal:

bash 
anythink entities create orders
anythink fields add orders status --type varchar
anythink fields add orders total --type decimal
anythink data create orders --data '{"status":"paid","total":40}'
anythink data create orders --data '{"status":"paid","total":25.5}'
anythink data create orders --data '{"status":"pending","total":10}'

Use a test project while you try this. The chat and MCP can change and delete records, and they run as you.

Step 1: Ask a question about your data

In the Anythink dashboard

  1. In the header, select the bot icon (Open AI panel).
  2. In the message box, type How many orders have status paid, and what do they total? and send it.
  3. The model calls a tool to read orders, then answers with the count and total.

The panel uses Anythink AI by default, which draws on your project's AI token allowance. To use your own Claude key instead, so the provider bills you and your allowance isn't used:

  1. Go to Settings, then AI Settings.
  2. Under AI Provider, select Bring Your Own Key.
  3. Under API Keys, paste your key into the Claude (Anthropic) row and select Save.
  4. In the AI panel, pick a Claude model from the model menu.

Note: Only Claude models call tools, so only Claude can look at or change your data. OpenAI, Grok and Gemini models chat without tools.

With the CLI

The CLI has no chat command. To see what the chat does for you, run the read it would run:

bash 
anythink data list orders --all --json

The chat reads the records in the same way, then counts the paid ones and adds up total itself.

With an AI assistant (MCP)

Once Claude is connected (step 3), ask:

text 
How many orders have status "paid", and what do they total?

Claude calls the cli tool with data list orders --all --json, counts the paid orders and adds up total itself.

Step 2: Have it change something

The chat acts as you, and a tool call runs as soon as the model makes it. The panel doesn't pause to ask first. Ask it to show you what it plans to do before it changes anything.

In the Anythink dashboard

In the AI panel, send:

text 
Which orders are pending? Don't change anything yet.

It lists order 3. Then send:

text 
Set order 3's status to paid.

The model updates the record and tells you it has. Open orders in the Anythink dashboard to see the new status.

With the CLI

The commands behind that exchange:

bash 
anythink data list orders --filter '{"status":"pending"}' --json
anythink data update orders 3 --data '{"status":"paid"}'

With an AI assistant (MCP)

Ask Claude the same thing:

text 
Which orders are pending? Don't change anything yet.

Claude runs data list orders --filter '{"status":"pending"}' --json. Then ask it to set order 3 to paid, and it runs:

text 
cli: data update orders 3 --data '{"status":"paid"}'
text 
āœ“ Record orders/3 updated.

Note: A change only succeeds if your role allows it. The chat and MCP have no permissions of their own, so a call you couldn't make through the CLI or REST API returns 403. For an assistant that should only look, connect with a key that has :read permissions only. See API keys.

Step 3: Connect Claude over MCP

MCP lets Claude Code, Claude Desktop or Cursor work on your project from outside the dashboard. The server runs on your machine, so there's nothing to add to the dashboard. This takes about two minutes. For other clients and every option, see Connect Claude to your backend with MCP.

In the Anythink dashboard

Open My Account, then API Keys, and create a key with only the permissions the assistant needs, for example orders:read. You can sign in with it in the next tab. Creating the key in the dashboard keeps it out of the conversation with the model.

With the CLI

  1. Install the CLI and the MCP server together, then sign in and pick your project:

    bash 
    brew install anythink-cloud/tap/anythink
anythink login
anythink accounts use <account-id>
anythink projects use <project-name>

  2. Register the server with Claude Code:

    bash 
    claude mcp add anythink anythink-mcp

To sign in with a key instead of a password, run anythink login --org-id <org-id> --api-key <api-key>.

With an AI assistant (MCP)

Start Claude Code and ask:

text 
Show me the active Anythink profile, then list the entities in this project.

Claude calls the config_show tool, then runs entities list. The profile's org id should match your project. If you haven't signed in with the CLI, ask Claude to sign you in with Google, which calls login_google and opens your browser. Then ask it to list your projects and pick one.

Check it worked

  • Dashboard: select Chat history in the AI panel. Your conversation is listed and reopens with your messages.
  • Change: open the orders records in the Anythink dashboard, or run anythink data get orders 3. Order 3 shows paid.
  • MCP: the config_show result lists your project as the active profile.
  • Answers: the model does the counting and adding itself, so check its maths on anything that matters.

Limits to know

Limit Detail
AI token allowance Anythink AI mode stops when your plan's allowance runs out, and the chat tells you to buy a top-up bundle or switch to your own key. Bring Your Own Key doesn't use it.
Claude only Only Claude models call tools. Other providers chat without access to your data.
Tool rounds The chat runs up to 10 tool rounds per message.
Commands in the dashboard chat A subset of CLI commands. MCP can run every CLI command.
MCP runs locally There's no hosted MCP endpoint to add as a remote connector.

See How AI chat and MCP work for the full list and Limits and quotas for plan quotas.

Next steps

AI assistant recipes for your backendSeven prompts to give Claude over MCP, from modelling an entity and fixing messy data to building a workflow and auditing access, with the commands it runs.
MCP server tools referenceEvery tool the Anythink MCP server gives Claude and other AI assistants, hosted and local, with arguments, sample calls and results.
How AI chat and MCP workHow the AI chat in the Anythink dashboard and the anythink-mcp server let Claude read and change your project's data, and where each one stops.