AI agent

CLI and MCP

Sheeta AI ships with a command-line tool named sheeta. It opens projects and files in the app, runs database commands with your project's connections, and starts the Sheeta MCP server for other AI tools. The in-app agent uses its own bundled copy, so you only need these steps to use Sheeta from a terminal or another AI tool.

Install the CLI

  1. Open Settings and select CLI.
  2. Click Install CLI. The status changes to CLI is installed.
  3. Restart your terminal so it finds the new command.

Installation may require administrator privileges on some systems. To remove the command, click Uninstall CLI.

Open projects and files

sheeta                      # open the app
sheeta .                    # open the current folder as a project
sheeta /path/to/folder      # open a folder as a project
sheeta /path/to/file.csv    # open the file's folder and then the file

A common workflow is to clone a repository with Git and open it right away:

git clone <repository-url>
sheeta <repository-folder>

Database commands

These commands use the connections saved in a project. Pass the project folder with --project-path and pick a connection with --connection.

sheeta tables --project-path /path/to/project --connection Local
sheeta describe orders --project-path /path/to/project --connection Local
sheeta query "SELECT * FROM orders LIMIT 10" --project-path /path/to/project --connection Local

tables and describe also accept --schema. Run sheeta <command> --help to see every option of a command. Results are printed as JSON.

The query command checks the project's AI permission rules before it runs SQL. It only runs a query that an Allow rule covers. A query that matches a Deny rule fails with Permission denied for this query, and a query with no matching Allow rule fails with Permission not configured. Add rules in the app under AI Permissions (see AI Agent).

Note: Settings > CLI in Sheeta AI 0.5.2 shows the flag as --project_path. The command expects --project-path.

MCP server

sheeta mcp starts a Model Context Protocol server over standard input and output. AI tools that support MCP can use it to list connections and tables, describe tables, run queries and manage dashboards in a Sheeta project. Credentials stay in the project and are never returned to the AI tool.

sheeta mcp --project-path /path/to/project

Without --project-path, every tool call must name the project folder itself.

Connect an AI tool from Settings

  1. Install the CLI as described above.
  2. Open Settings and select MCP.
  3. Find your tool in the list: Claude Code, Cursor IDE or Gemini CLI. Tools that are not installed show Not detected on this system.
  4. Click Add MCP Server. The status changes to Configured. Click Remove to undo it.

What Sheeta AI changes for each tool:

  • Claude Code: runs claude mcp add to add a user-level server named sheeta.
  • Cursor: adds a sheeta entry to ~/.cursor/mcp.json.
  • Gemini CLI: adds a sheeta entry to ~/.gemini/settings.json.

Manual configuration

For any other MCP client, add a server that runs the sheeta command with the mcp argument. Settings > MCP shows this snippet with the real path of the command on your computer:

{
  "mcpServers": {
    "sheeta": {
      "command": "/path/to/sheeta",
      "args": ["mcp"]
    }
  }
}

Deep links

Sheeta AI also opens sheeta:// links. Every link needs a project parameter with the project folder path.

  • sheeta://query?project=<path>&conn=<name>&sql=<sql> opens a new query tab with the SQL.
  • sheeta://table?project=<path>&conn=<name>&schema=<schema>&table=<table>&view=data opens a table. Use view=structure for the structure view.
  • sheeta://dashboard?project=<path>&name=<name> opens a dashboard by name.

Values in a link must be URL-encoded. See Troubleshooting if a command or integration does not work.