Set up the Rastro MCP server to run catalog operations from your AI agent. Pull snapshots, transform data locally, stage changes for review, and apply from the dashboard.
No database credentials needed — the MCP server talks only to the Rastro API.
Prerequisites Authenticate This opens dashboard.rastro.ai in your browser. Log in, and your credentials are saved automatically.
Claude Code
Claude Desktop
Cursor
Codex
claude mcp add rastro -- uvx rastro.ai
If using an API key instead of browser login: claude mcp add rastro \ -e RASTRO_API_KEY=rastro_pk_... \ -- uvx rastro.ai
Open your config file:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.jsonWindows: %APPDATA%\Claude\claude_desktop_config.json If you ran uvx rastro.ai login above: { "mcpServers" : { "rastro" : { "command" : "uvx" , "args" : [ "rastro.ai" ] } } }
Or with an API key: { "mcpServers" : { "rastro" : { "command" : "uvx" , "args" : [ "rastro.ai" ], "env" : { "RASTRO_API_KEY" : "rastro_pk_..." } } } }
Restart Claude Desktop after saving. Open Cursor Settings (Cmd/Ctrl + Shift + J) → MCP tab → Add new MCP server . Or add to .cursor/mcp.json in your project root: { "mcpServers" : { "rastro" : { "command" : "uvx" , "args" : [ "rastro.ai" ] } } }
If using an API key instead of browser login, add an "env" field: "env" : { "RASTRO_API_KEY" : "rastro_pk_..." }
Add to ~/.codex/config.toml: [ mcp_servers . rastro ] command = "uvx" args = [ "rastro.ai" ]
If using an API key instead of browser login: [ mcp_servers . rastro ] command = "uvx" args = [ "rastro.ai" ] env = { RASTRO_API_KEY = "rastro_pk_..." }
Verify Ask your agent:
If the server is connected, it will call catalog_list and return your catalogs.
First workflow Prompt your agent:
For catalog <catalog_id>, pull a snapshot, update price/cost by my rules, compute diff, validate, and stage one pending-review activity. Do not apply automatically.
Expected flow:
execution_catalog_snapshot_pullLocal transform (agent-generated Python)
execution_local_diff_computeexecution_bundle_validatecatalog_activity_create_transform (or execution_catalog_stage_dataset)Review and apply in the dashboard
Safety defaults
Programmatic approve/apply is blocked — all changes require dashboard review.
Large change sets are staged into one activity (chunked internally).
Always review the dashboard diff before applying.
Next
MCP Reference Full tool reference and troubleshooting
Catalog API Underlying API endpoints
