Connect Claude to your backend with MCP

anythink-mcp is a Model Context Protocol server for Anythink. Your MCP client launches it as a local process and talks to it over stdio. Once it's connected, Claude can create entities and fields, read and write records, and build and debug workflows in your project. For how it relates to the AI panel in the Anythink dashboard, see How AI chat and MCP work.

Install it

The server and the CLI are both .NET global tools, so you need the .NET SDK:

dotnet tool install -g anythink-cli
dotnet tool install -g anythink-mcp

Install the CLI too. The server's cli tool runs your installed anythink binary, and without it every cli call fails with a "CLI not found" error.

Register it with your client

Claude Code

claude mcp add anythink anythink-mcp

Or add it to your project's .mcp.json:

{
  "mcpServers": {
    "anythink": {
      "command": "anythink-mcp"
    }
  }
}

To pin a saved CLI profile instead of whichever one is active, pass --profile:

{
  "mcpServers": {
    "anythink": {
      "command": "anythink-mcp",
      "args": ["--profile", "my-project"]
    }
  }
}

Claude Desktop and other clients

Add the same mcpServers entry to the client's MCP configuration file. Any client that can launch a stdio server works.

Sign in and pick a project

The server shares its saved profiles with the anythink CLI, so you can sign in either way.

With the CLI

This keeps your password out of the conversation with the model:

anythink login
anythink accounts use <account-id>
anythink projects use <project-name>

anythink login --google signs in with Google instead. The MCP server picks up the new profile on its next tool call.

With an AI assistant (MCP)

Sign me in to Anythink with Google, then connect to my "production" project.

Claude calls:

1. login_google, which opens your browser to approve the sign-in. The login tool takes an email and password instead.
2. accounts_list, then accounts_use with your billing account's id.
3. projects_list, then projects_use with the project's name, org id or id prefix.

projects_use saves a profile for that project and makes it active. Unless you pass an apiKey, it generates a project-scoped access token. Every later cli call uses that profile.

Connect with an API key instead

If you already have a project's org id and an API key, you can skip the billing account step. Run this in a terminal, so the key never passes through the model:

anythink login --org-id <org-id> --api-key <api-key>

The MCP equivalent is the login_direct tool with orgId and apiKey. It saves a profile named after the org id (or profile, if you pass one) and makes it active.

What the assistant can do

The cli tool runs any anythink command against the active profile: entities, fields, data, workflows, roles, users, secrets, files, integrations, menus and raw API calls through fetch. For every tool and its arguments, see MCP server tools reference.

The assistant can make destructive changes, including deleting entities, records and projects. Read what it plans to run before you approve a write, and use a scoped role when you only need it to read.

Permissions

MCP calls have no permissions of their own. They run as the signed-in user or the API key's role. For an assistant that can look but not change anything, connect with an API key whose role has only :read permissions on what it needs, such as anythink_entities:read and anythink_workflows:read plus read access to your own entities. See Roles and permissions.

Example prompts

List every entity in this project and the fields on each.

Create an entity called "feedback" with a text field "message" and an
integer field "rating", then add three sample records.

Show me the recent jobs for the "Flag large orders" workflow and tell me
why the failed ones failed.

Limits

Limit: The cli tool rejects commands that contain query strings (?, &), currency symbols or most punctuation beyond - . / : = , @ " ' { } [ ]. Keep JSON payloads to plain text, and don't put apostrophes inside single-quoted --data values.

Note: You can't point an MCP client at Anythink's hosted server. The copy that powers the dashboard's AI panel isn't a public MCP endpoint, so always run anythink-mcp locally.

Next steps