# Connect Claude Source: https://www.luriart.com/docs/start/claude Plain text: https://www.luriart.com/docs/start/claude.txt Let Claude (or any MCP agent) set Luria up and run your split tests. One connection, one pasted prompt. Luria is an MCP server at `https://www.luriart.com/api/mcp`. Connect it once, paste one prompt, and Claude does the setup: adds your site, checks the tag, starts the first big test, reads results. ## Connect **Claude Code** ```bash claude mcp add --transport http luria https://www.luriart.com/api/mcp ``` Then type `/mcp`, pick **luria**, and sign in with your email. Press **Allow**. **Claude.ai or Claude Desktop**: Settings, Connectors, Add custom connector. Name `Luria`, URL `https://www.luriart.com/api/mcp`. Sign in when asked and press **Allow**. **Cursor, ChatGPT, Codex**: add a remote MCP server with the same URL. The sign-in is the same. No key to copy: signing in gives the agent a key you can revoke in the dashboard under **Settings, API key**. ## Paste one of these The agent calls `setup_status` first. Luria answers with your tier and the next steps, so the prompt stays short. **First time (no account yet)** ```text Connect the Luria MCP (https://www.luriart.com/api/mcp). If it asks, sign me in with my email and I will press Allow. Then call setup_status and do every step it lists: add my site https://YOURSITE.COM with goal Sales, give me the tag to paste before , run verify_install until it is live, then auto_test so Luria starts the first big test. Call setup_status again until it lists nothing new, then give me the dashboard link. ``` **Free (you have an account, on the 20 free split tests)** ```text Connect the Luria MCP (https://www.luriart.com/api/mcp), sign me in if asked. Call setup_status and do what it lists for each of my sites: paste or verify the tag, start a test with auto_test where nothing runs, and tell me how many free split tests are left. Then show list_tests with each test's state in one line each. ``` **Paid** ```text Connect the Luria MCP (https://www.luriart.com/api/mcp), sign me in if asked. Call setup_status, fix anything it flags (tag, no test running), then read every running test with get_test and tell me in one line each: who is ahead, how sure, and whether to promote with change_test. Promote only what it calls a winner. ``` Replace `YOURSITE.COM`. Every tool asks nothing it cannot do: a price test still waits for your tick in the dashboard. ## What the agent can call `setup_status`, `add_site`, `site_status`, `verify_install`, `auto_test`, `create_test`, `list_tests`, `get_test`, `change_test`, `record_conversion`, `credits`, `buy_credits`, `dashboard_link`, plus the personal-pages tools. The same calls exist as REST at [luriart.com/llms.txt](https://www.luriart.com/llms.txt).