connection_setup
ChatGPTGenerate a browser URL for configuring a connection in the web app.
Overview
MarcoPolo spins up a secure container where ChatGPT can work with your actual data. Connect to your databases, APIs, S3, lakehouses, CRMs, Jira, logs and much more—using scoped credentials that are never exposed to the model. ChatGPT gets DuckDB, Python, a shell, and a set of tools to explore, query, transform, and analyze data across systems. The workspace persists over time, so that you can build on your work. Prep a report, understand your data, debug an issue, or review the latest metrics right here in the conversation.
Tools
connections_list
ChatGPTList visible Marcopolo connections ready in the execution workspace. Use connection.name with the connection CLI through workspace_shell. Entries with connection.queryable=false need provider authorization; follow their connection.next_step and note. Each entry lists its capabilities; only run queries against a connection whose capabilities include 'query'.
data_query
ChatGPTRun a bounded governed query against a visible connection. Use this tool in generated code that re-queries live data at view or load time: Remote Artifacts, external web apps, scheduled scripts, and any other programmatic interface that fetches fresh data on each run. Do NOT use it for agent-driven queries or one-off snapshot visualizations — use the
workspace_shell tool with the connection query CLI instead. Pass the full workspace-relative query_file path including the connection prefix, e.g. connections/<name>/queries/<file>. Paths resolve from /workspace, not from any current directory. Author new query files through workspace_shell before calling this tool so query text is durable and reusable. Returns rows as a parsed list of record dicts and row_count as the total — no JSON parsing needed. Results are capped by max_rows (1–5000).install_demo_connection
ChatGPTInstall a hosted demo connection by id or natural-language intent. If the request is ambiguous, the tool returns matching demo connection options.
preview_dashboard
ChatGPTPreview a dashboard manifest. The path should point to a .dashboard file.
workspace_shell
ChatGPTExecute a command in the user's remote workspace. This is a remote Linux workspace, not the user's local machine. Work in
/workspace like a checked-out repo: - read /workspace/README.md, /workspace/RULES.md, and /workspace/workflows/README.md - inspect files with normal shell tools such as ls, find, rg, grep, cat, and sed - create and edit /workspace/... files with shell commands in this remote workspace - use git status and git diff as part of normal authored work - run connection list --json before connection work and use each connection's capabilities as the authoritative command surface - use browse, download, or upload only when that verb is listed in the selected connection's capabilities - use cron create <name> --command "<workspace command>" --cron "<expr>" for managed recurring jobs in the workspace - write query files under /workspace/connections/<name>/queries/ Query and browse results can be materialized into the workspace DuckDB state stored under .dv/duckdb/workspace.duckdb; use connection query DUCKDB --file <query-file> --json to work with that local analytical state.Capabilities
Writes
Example Prompts
Click any prompt to copy it.
