CommunityHire an ExpertSign Up
Softr APIApp MCP server

App MCP Server

Let the users of your Softr app work with it through their own AI assistant, with the exact same permissions they have in the app.

Every Softr app can act as its own Model Context Protocol (MCP) server. Once you enable MCP on an app, the people who use that app can connect their own AI assistant (Claude, ChatGPT, Mistral, and others) and work with the app through it.

It turns your app into a surface an assistant can operate on behalf of the user. Whatever a user does by clicking through the app’s screens, reading the records they have access to, submitting a form, updating an entry, they can now ask their assistant to do in plain language instead. Every request runs as that specific user, so user groups, page visibility, and data restrictions all still apply, and the MCP server can never expose more than the user’s own account already can. That makes it a safe and reliable way to bring AI into your app.

For example, if you’ve built a CRM or an ERP on Softr, a user could ask their assistant to check the stock left on a product, adjust a quantity, or add a new contact, and each action respects the same permissions that user has in the app itself.

How permissions work

When an AI client connects, it does so as a signed-in user of your app, not as an admin and not as the app itself. Everything the assistant can see or do is capped by what that user could already do in the app’s interface. The MCP server never grants extra access.

Concretely, the assistant’s capabilities are derived from the blocks you’ve built into the app:

What the app containsWhat the assistant can do
A list, grid, or table blockRead records
A form or sign-up blockCreate records (only the fields shown in the form)
A profile block or inline editingUpdate records
A delete actionDelete records

This means the MCP surface mirrors your app’s screens:

  • If there’s no block for a table, the assistant can’t touch that table, even if the underlying data source contains it.
  • Fields are limited to those your blocks display. Hidden or unmapped fields stay invisible to the assistant.
  • User groups, page visibility, and data restrictions all still apply. A user connecting through MCP sees exactly the records and actions their account would see when logged into the app normally.

Enabling MCP for your app

Open the App as MCP settings

In the app’s left sidebar, open Settings → App as MCP.

Enable MCP

Click Enable MCP. Softr generates an OAuth Client ID and Client Secret for this app.

Connect an AI tool

Use one of the quick-setup cards (Claude, Claude Code, ChatGPT, Other) to connect a client. Each card walks you through the steps for that specific tool.

Server URL

Your app’s MCP endpoint follows your app’s published URL:

https://<your-app-domain>/v1/mcp

For example, if your app is published at https://acme.softr.app, the MCP server URL is https://acme.softr.app/v1/mcp. If you’ve added a custom domain, the MCP URL uses that domain. You can copy the exact URL from the Server URL field in the App as MCP settings.

Connecting an AI tool

For most clients, the Server URL is the only thing you need. You paste it into the client, and the client opens a browser window where the user signs in to the app and approves the connection over OAuth. Nothing else has to be shared. The Client ID and Secret in Advanced settings are only needed for the few clients that can’t register themselves automatically.

The setup dialog in the App as MCP settings gives you a ready-to-use snippet for each supported client. You can also follow the steps manually.

Open the connector setup

In Claude, go to Customize → Connectors → + → Add custom connector.

Add the server

Enter a name (for example, your app’s name) and paste the Server URL from the App as MCP settings.

Authorize

Click Add. Claude opens a browser window to authorize the connection to your app via OAuth.

Claude uses dynamic client registration, so you don’t need to paste a Client ID or Secret.

Claude Code is Anthropic’s CLI. The setup snippet looks like this:

Terminal
claude mcp add <slug> https://<your-app-domain>/v1/mcp --transport http --client-id <your-client-id> --client-secret

Install Claude Code

Terminal
npm install -g @anthropic-ai/claude-code

Run the command

Copy the command from the Claude Code setup card in the App as MCP settings and run it in your terminal. The Client ID is already filled in.

Paste the Client Secret when prompted

Claude Code’s interactive CLI displays:

Enter OAuth client secret:

Paste the Client Secret shown in the dialog (or in Advanced settings → Client Secret) and press Enter. The CLI deliberately prompts interactively so your secret doesn’t end up in your shell history.

Authorize

After submitting the secret, Claude Code opens a browser window to complete OAuth authorization.

Enable Developer Mode

In ChatGPT, go to Settings → Connectors → Advanced and enable Developer Mode.

Create a custom connector

Open Settings → Connectors → Create. Enter a name, paste the Server URL from the App as MCP settings, and leave the Authentication options as-is.

Connect

Click Create, then Connect. ChatGPT opens a browser window to authorize access to your app.

Any OAuth-capable MCP client can connect using just the Server URL. Most clients perform dynamic client registration automatically and don’t need a pre-shared Client ID.

JSON
{
  "url": "https://<your-app-domain>/v1/mcp",
  "transport": "http"
}

If your client requires a pre-registered Client ID and Secret, copy them from Advanced settings.

Advanced settings

Open the Advanced settings section in the App as MCP settings to see and manage the OAuth credentials.

FieldWhat it’s for
Client IDPublic identifier for your app’s MCP OAuth client. Safe to share.
Client SecretConfidential credential. Used by clients that don’t support dynamic client registration.
Redirect URIsAllowed OAuth redirect URIs. Add an entry for every client that needs a fixed redirect. Most clients fill this in automatically via dynamic client registration.

Adding redirect URIs

Some AI clients require their callback URL to be pre-registered on the OAuth server. To add one:

Open Advanced settings

Expand Advanced settings in the App as MCP settings.

Add the URI

Click Add redirect URI and paste the callback URL the client provided.

Turning MCP off

Toggle Enable MCP off and click Save to disable the server without removing its credentials. Existing connected clients will stop being able to call the server until you re-enable it.

Troubleshooting

My client still asks for the Client Secret after I pasted it inline

Some CLIs (Claude Code included) treat --client-secret as a prompt-trigger flag, not a value flag. Any value you append inline is ignored and the CLI prompts interactively. This is intentional so the secret doesn’t appear in your shell history. Paste the secret at the prompt instead.

OAuth authorization redirects to an error

Check that the client’s redirect URI is registered under Advanced settings → Redirect URIs. Clients that support dynamic client registration handle this automatically; clients with a fixed callback don’t.

Use these docs with your AI tools

An AI agent can read this documentation directly. You do not need an account or an API key. Everything here is public and read-only.

Query these docs via MCP

Recommended

Add this server to Claude, Claude Code, Cursor, Mistral, or any tool that supports MCP. Your agent can then search Softr Documentation documentation and read it in full, instead of answering from memory.

https://docs.softr.io/mcp
  • searchFind the passages that answer a question.
  • fetchRead one page in full, as Markdown.
  • list_pagesSee every page in this documentation.

Query these docs over HTTP

The same tools also work as plain web requests. Use this for scripts, or for any tool that does not support MCP. There is one endpoint per tool. Arguments go in the query string, and the answer comes back as JSON.

https://docs.softr.io/api/docs/search?query=custom+domain

Read the OpenAPI description. It is built from the same definitions as the tools, so it always matches what the endpoints do.

Read these docs as Markdown

Add .md to any page URL to get its Markdown source. You can also send the headerAccept: text/markdown to the page URL itself.

To read the whole documentation in one file, open llms-full.txt. For a short index of every page, open llms.txt.