MCP Server

Data Island includes a built-in Model Context Protocol (MCP) server that lets AI assistants like Claude, ChatGPT, and Copilot interact with your data directly. Ask questions in natural language and get back SQL-powered answers.

What is MCP?

The Model Context Protocol is an open standard that allows AI assistants to connect to external data sources and tools. Instead of copying data into a chat window, MCP gives the AI direct, permission-aware access to query your tables, explore schemas, and store findings.

Open source: MCP is an open protocol maintained by Anthropic. Any AI assistant that supports MCP can connect to Data Island.

Setup with Claude Desktop

Add the Data Island MCP server to your Claude Desktop configuration file.

macOS

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

claude_desktop_config.json
{
  "mcpServers": {
    "dataisland": {
      "command": "supertable",
      "args": ["mcp", "start"],
      "env": {
        "SUPERTABLE_ORGANIZATION": "my-org",
        "SUPERTABLE_SUPERUSER_TOKEN": "your-token",
        "SUPERTABLE_API_URL": "http://localhost:8050"
      }
    }
  }
}

Windows

Edit %APPDATA%\Claude\claude_desktop_config.json with the same configuration.

After saving the config, restart Claude Desktop. You should see the Data Island tools appear in the tool picker.

Available Tools

The MCP server exposes 24 tools organized into four categories.

Discovery (6 tools)

Explore your data landscape without writing any code.

Tool Description
list_tables List all available tables with row counts and metadata
describe_table Get full schema with column names, types, and descriptions
get_table_stats Retrieve storage size, partition count, and version info
sample_data Preview a random sample of rows from a table
list_versions Show the version history of a table
search_columns Find columns matching a name or type pattern across all tables

Query (6 tools)

Run SQL, profile data, and get type-aware results.

Tool Description
execute_query Run a SQL query and return results as a table
explain_query Show the query execution plan
profile_column Get min, max, mean, nulls, unique count for a column
count_rows Fast row count for a table or filtered subset
time_travel_query Query data as of a specific timestamp or version
compare_versions Diff two versions of a table to see what changed

Feedback (6 tools)

Store analysis results and annotate findings for collaboration.

Tool Description
store_finding Save an analysis finding with title, description, and evidence
list_findings Retrieve previously stored findings
annotate_column Add a description or tag to a column
annotate_table Add a description or tag to a table
flag_quality_issue Report a data quality issue for a table or column
suggest_query Get SQL query suggestions based on the table schema

State (6 tools)

Persist conversation context across sessions.

Tool Description
save_context Persist the current analysis context for future sessions
load_context Restore a previously saved context
list_contexts List all saved analysis contexts
delete_context Remove a saved context
save_query_bookmark Bookmark a useful query for reuse
list_bookmarks List all saved query bookmarks

Example Conversation

Here is what interacting with Data Island through an AI assistant looks like.

User:

What tables do I have, and which one has the most rows?

Claude:

I'll check your tables using the Data Island MCP tools.

Used tool: list_tables
You have 5 tables. The largest is events with 2.4 million rows, followed by sensors with 840,000 rows.

User:

Show me a breakdown of event types from last week.

Claude:

Used tool: execute_query with SQL:
SELECT event_type, COUNT(*) FROM events WHERE timestamp > NOW() - INTERVAL '7 days' GROUP BY event_type ORDER BY 2 DESC

Security

The MCP server respects all RBAC policies configured in Gatekeeper. The AI assistant can only access tables and operations that the configured token allows. No data leaves your infrastructure.

Important: Use a scoped API key with minimal permissions for the MCP server, not the superuser token. Grant read-only access unless the assistant needs to write data.

Next Steps