MCP Integration

Connect AI agents like Claude, Cursor, and Windsurf to your Rowmesh app data using the Model Context Protocol (MCP)

New to MCP? Start with Connect your own AI agent — it walks through both the app-data MCP and the guides MCP together.

The Rowmesh MCP (Model Context Protocol) server lets AI assistants list your tables, read records, and create, update, or delete entries through natural language.

One MCP URL covers every Rowmesh app where you're a data admin:

https://api.core.rowmesh.com/mcp

There's no API key and no per-app URL. The first time a client calls the server, a browser window opens for you to sign in and pick which apps the client can access and whether it gets read or read + write on each.

The Rowmesh consent screen — pick which apps the client can use, and whether it gets read or write on each.

A connected client acts with your data admin role on every app you approve. It can read and modify all records in those apps regardless of row-level permissions. Only approve clients you trust.

Connecting an AI Client

The Rowmesh MCP server uses with OAuth, which all major MCP clients support. Pick your client:

Claude Desktop

One-click connector link

Claude Code

A single CLI command

Cursor & Windsurf

Paste one JSON block

ChatGPT

Turn on developer mode, then create a plugin

Other MCP clients

Anything supporting OAuth over Streamable HTTP

Claude Desktop

The quickest route is one click: . That link opens the Rowmesh connector in Claude's directory — click Connect, and the Rowmesh consent screen opens in your browser.

You can also find it from Settings → Connectors: browse the directory, search for Rowmesh, and click Connect.

If your organization restricts the connector directory, you can still add Rowmesh as a custom connector: click Add custom connector, enter a name (e.g. Rowmesh) and https://api.core.rowmesh.com/mcp as the remote MCP server URL, then click Add. Free Claude plans are limited to one custom connector.

Claude Code

Run:

claude mcp add --transport http rowmesh https://api.core.rowmesh.com/mcp

The next time you start a session, Claude Code opens the consent screen in your browser.

Cursor & Windsurf

Open the MCP settings panel and add:

{
  "mcpServers": {
    "rowmesh": {
      "url": "https://api.core.rowmesh.com/mcp"
    }
  }
}

The first tool call opens the consent screen.

ChatGPT

ChatGPT has no one-click link, so there are three short steps: check your plan, turn on developer mode, then create the plugin.

1. Check your plan

Custom MCP plugins need a Plus, Pro, Business, Enterprise, or Education plan, and you have to set them up in the ChatGPT web app — the mobile apps can't create them. Once created, the plugin works everywhere you use ChatGPT.

2. Turn on developer mode

In the ChatGPT web app, go to Settings → Apps → Advanced settings and switch on Developer mode.

On a Business or Enterprise workspace? If you don't see the Developer mode toggle, an admin has to allow it first in Workspace Settings → Permissions & Roles → Connected Data → Create custom MCP connectors.

3. Create the plugin

Developer mode adds the option to create your own plugin. You're in the right place when the dialog is titled New Plugin. Fill it in like this:

Field	What to enter
Icon	Optional — skip it. PNG only, 256 × 256 px or larger, max 10 KB.
Name	Rowmesh
Description	Access to my Rowmesh app data. Use this to list my Rowmesh projects and their tables, read table schemas, and read, create, update, or delete records in those tables.
Connection	Leave it on Server URL — not Tunnel — and enter https://api.core.rowmesh.com/mcp, including the /mcp path
Authentication	OAuth

Although the Description is marked optional, it's worth filling in. ChatGPT reads it when deciding whether a question calls for Rowmesh, so a vague one means the model skips the plugin when you wanted it. Feel free to make it more specific to your own apps, for example by naming the tables you query most.

You can leave Advanced OAuth settings alone. ChatGPT discovers our OAuth details from the server URL and registers itself through dynamic client registration, so there's no client ID or secret to paste — open that panel and you'll see the discovered settings already filled in.

Save the plugin. The first time you use it, the Rowmesh consent screen opens for you to sign in and pick which apps ChatGPT can access.

ChatGPT will warn you that custom plugins aren't developed or verified by OpenAI. That's expected for any custom MCP server, including this one — the warning is about who built it, not about whether this URL is safe. Check that the server URL reads exactly https://api.core.rowmesh.com/mcp before you approve it.

Other MCP Clients

Any client that supports OAuth over Streamable HTTP can connect. Point it at https://api.core.rowmesh.com/mcp — no header, no key. The client handles the OAuth flow on first connection.

Available Tools

Every data tool takes a projectName argument. The agent calls list_projects first to discover which apps it can see, then passes the right project name on each call.

Tool	What it does
list_projects	Lists every Rowmesh app you approved for this client, with the scope (read or read + write) granted on each
list_tables	Lists tables in an app
get_table_schema	Returns a table's field types, options, and relationships
list_records	Fetches records with optional filtering, sorting, and pagination
get_record	Fetches a single record by ID
create_record	Creates a record
update_record	Updates a record by ID
delete_record	Deletes a record by ID

Security and Access Management

• OAuth, not API keys. No long-lived secret is stored in the client config.

• Per-app role check on every call. If you lose data admin access on an app, the next call from any connected client is rejected — no token reissue needed.

• Per-client revocation. Each client has its own OAuth token; revoking one doesn't affect the others.

• Scope chosen at sign-in. Pick read-only or read + write per app on the consent screen. Re-run the flow any time to change either.

• HTTPS only.

FAQ

MCP FAQ