Anythink gives AI models access to your project in two ways: an AI chat panel in the Anythink dashboard that answers questions and makes changes through tools, and an MCP server (anythink-mcp) that lets Claude Code, Claude Desktop or another Model Context Protocol client work on your project from outside the dashboard.

Both use the same set of tools, and both act with your permissions. For calling a model as a step inside an automation, see the Integration step in Step types.

The moving parts

Part	What it is
AI chat panel	The panel you open from the bot icon in the dashboard's header. It streams a reply and can call tools mid-conversation to read or change data.
Provider	The model vendor behind a chat: Claude (Anthropic), OpenAI or Grok (xAI).
AI mode	Anythink AI (platform-managed, uses your credit balance) or Bring Your Own Key (your provider key, billed by the provider).
Connection	A provider API key stored for your project, encrypted at rest. Added in Settings › AI Settings or Settings › Integrations.
MCP server	`anythink-mcp`, a .NET tool you install and run locally. Your MCP client launches it and talks to it over stdio.
MCP tool	One capability the server exposes: signing in, choosing a project, or running an Anythink CLI command. See MCP server tools reference.

How a chat message runs

You send a message in the AI panel, optionally with image or PDF attachments.
Anythink picks the provider from the model name. Models starting gpt-, o1, o3 or o4 go to OpenAI, grok- models go to Grok, gemini- models go to Gemini, and everything else (including all Claude models) goes to Claude.
It picks the credentials. In Bring Your Own Key mode, Anythink uses your enabled connection for that provider. In Anythink AI mode, or if you have no connection for that provider, it uses the platform key and checks your credit balance first.
It runs the conversation loop. The model replies; if it asks for a tool, Anythink runs the tool as you and feeds the result back. This repeats until the model stops asking for tools, up to 10 rounds.
The reply streams back as it's generated, with a marker each time a tool runs.
The conversation is saved so you can reopen it from Chat history in the panel.

Before the first round, Anythink gives the model a short summary of your project: how many entities it has and the names of the 10 most recently created ones. Anything more, it has to fetch with a tool.

Note: Only Claude models call tools. OpenAI and Grok models chat without tools, so they can't look at your data.

Anythink AI vs Bring Your Own Key

Choose the mode in Settings › AI Settings › AI Provider.

Mode	Behaviour
Anythink AI	Uses Anythink's credentials and your credit balance. Pick the model under Default Model. When the balance runs out, the chat stops with a message telling you to top up or switch to your own key.
Bring Your Own Key	Uses the key you save under API Keys for the provider of the model you pick in the panel. Billed by the provider, with no credit usage.

In Bring Your Own Key mode, the panel only lists models that work with your own keys. If you pick a model whose provider has no saved key, Anythink falls back to the platform key and your credit balance, which only works for Claude models. To use OpenAI or Grok in chat, save a key for that provider first.

Note: AI Settings lists Gemini and a saved Gemini key does work in the panel, but the API Keys section's Connected badge only checks for Claude, OpenAI and Grok connections — a saved Gemini connection never shows as connected there, even though chat requests to gemini- models use it.

The same Claude, OpenAI and Grok connections power the Integration workflow step. See How integrations work and Integration providers reference.

The MCP server

anythink-mcp is published from the public anythink-cli repository as a .NET global tool. Your MCP client starts it as a local process. It reads the same saved profiles as the anythink CLI, and its cli tool runs your installed anythink binary. That means the assistant can run any CLI command you could. To set it up, see Connect Claude to your backend with MCP.

The dashboard AI panel reaches the same tools through a copy of the server that Anythink hosts. It supports a subset of CLI commands (listed in MCP server tools reference), and it isn't an MCP endpoint you can connect your own client to.

The server has 16 tools: five for signing in and out, three for local profiles, three for billing accounts, four for projects, and a cli tool that runs Anythink CLI commands.

Permissions

The AI panel and the MCP server act as you (or as the API key you connected with). There's no separate AI or MCP permission: a tool call succeeds only if your role allows the same action through the CLI or API.

To	Permission
Use the AI chat panel	`anythink_ai_chat:create`
See available models and your chat history	`anythink_ai_chat:read`
Read or change the project's AI mode and default model	`anythink_ai_config:read` / `anythink_ai_config:update`
Save or replace a provider key in AI Settings	`anythink_integrations:read`, `:create` and `:delete`
Run a one-off prompt with `anythink integrations execute`	`anythink_integrations:read`

Tool calls need the permission for what they touch. Listing entities needs anythink_entities:read; creating a workflow needs anythink_workflows:create.

Limits and trade-offs

Limit	Detail
Tool rounds per message	10. If the model is still calling tools after that, the reply ends where it is.
Providers that use tools	Claude only. OpenAI and Grok chat without tools.
Gemini	Works in chat, but a saved Gemini key never shows as Connected on the AI Settings page — that badge only checks Claude, OpenAI and Grok connections.
Attachments	JPEG, PNG, GIF, WebP or PDF, up to 5 MB each and 5 per message.
`anythink integrations execute`	Claude only. OpenAI and Grok aren't supported by this command yet; use the Integration workflow step for them.
Commands in the dashboard panel	A subset of CLI commands. Integrations, menus, payments, migrations and search only work through a local MCP server.
Characters in `cli` commands	Commands can't contain query strings (`?`, `&`), currency symbols or most punctuation beyond `- . / : = , @ " ' { } [ ]`.
Chat history	Saved per user, per project. You see only your own conversations.

Next steps

Chat with your backend data using AI

Connect Claude to your backend with MCP

Install the anythink-mcp server and connect Claude Code, Claude Desktop or another MCP client to your Anythink project to read and change its data.

MCP server tools reference AI assistant recipes for your backend

How AI chat and MCP work