Skip to main content
Model Context Protocol (MCP) is an open standard that lets AI assistants securely connect to external tools and data. Softr’s MCP server gives an assistant like Claude, ChatGPT, or Cursor direct access to your Softr workspace — so you can query and edit your data, read and build apps, write custom code blocks, and create workflows, all in plain language. It is the same workspace you see in Softr Studio, reached a different way. Anything the assistant does happens as you, with your permissions, and shows up in the builder like any other change.

​
What you can do

​
How Softr is organized

Almost every tool asks for a workspace, so it helps to know how the pieces fit together. A workspace is the top-level container, and everything else belongs to one:
list_workspaces returns the workspaces you can reach, with their IDs. It is how the assistant turns “my Sales workspace” into the ID that every other tool needs, so it is often the first call in a conversation.

​
MCP server URL

Use this URL when adding Softr as an MCP server in your AI tool.

​
Authentication

Softr’s MCP server supports two ways to sign in. Both let you choose exactly what the AI tool can reach. OAuth is the normal route, and what Claude, Claude Code, ChatGPT, Codex, Cursor and most other clients use. You give the client the MCP server URL, and Softr shows you an authorization screen where you pick what to grant. Most clients negotiate the OAuth details themselves — a Client ID only has to be entered by hand in clients that cannot register with Softr on their own, such as Cursor. A personal access token covers custom clients, your own MCP build, and any tool with no OAuth support. Create one under API tokens in your account menu. Connect your AI assistant has the exact steps for every tool, the Client IDs for the clients that ask for one, and what the authorization screen is granting.

​
Permissions

Whether you connect with OAuth or a personal access token, you choose permissions the same way: pick which workspaces are covered, then set an access level for each of three areas. What each level allows:
Levels are bundles, not a checklist. There is no way to grant editing without reading, or database schema changes without record access. Picking a level always includes everything below it.Deleting travels with editing, too: Edit data lets an assistant delete records, and Full access lets it delete fields, tables, and whole databases. If you want a token that can never delete anything, the level to pick is View only. See Deleting data for the rules that still apply.

​
Managing access

Go to API tokens in your account menu to see and remove access at any time.
  • Personal access tokens can be edited, regenerated, or deleted.
  • Authorized apps lists the AI tools you connected with OAuth, with a Revoke button for each. Revoking one immediately invalidates the tokens it was issued.
The Authorized apps section only appears once you have authorized at least one app, so it will be missing if you have only ever used personal access tokens.

​
Example prompts

Once connected, you can work with your workspace in plain language:
  • “Create a new blank app in my workspace.”
  • “Add a user named Jane Doe with email jane@example.com to my app.”
  • “Deactivate the user jane@example.com in my app.”
  • “List my app’s users — who hasn’t signed in yet?”
  • “Disable login for my app.”
  • “Only let people with an example.com email sign up for my app.”
  • “Send my app’s emails from support@acme.com as Acme Support.”
  • “What email sender is configured on my app?”
  • “List all databases in my workspace.”
  • “Search the Orders table for everything still pending and sort by date.”
  • “Add a Priority field to Tasks with options Low, Medium, and High.”
  • “Build me a chart block on the dashboard page showing revenue by month.”
  • “What user groups can edit the Clients page?”
  • “Create a workflow that sends me a Slack message whenever a new order comes in.”
  • “Show me what the app looks like right now.”