# What to ask your agent > Example prompts for Claude Code, Cursor or Codex CLI once GetIntel is connected over MCP, from checking where you stand to shipping the fix in your codebase. Source: https://getintel.ai/docs/mcp/example-prompts/ Once GetIntel is [connected](/docs/mcp/connect/), talk to your agent in plain language. It picks the right tools; the names in brackets show which ones it will usually call. ## Where do I stand? - "Give me this week's AI visibility summary and what changed." (`get_overview`) - "Which of my topics are weakest, and who leads each one?" (`get_topics`) - "Which buyer questions did I lose since last week?" (`get_prompt_history`) - "How many MCP reads have I used this month?" (`get_usage`) ## Why is it happening? - "Open the pricing topic. Which questions am I missing and what does ChatGPT say instead?" (`get_topic`, `get_receipts`) - "Why does AI pick Competitor X over us? Show the topics they win." (`get_competitor`) - "Which domains does AI cite in our category that never mention us?" (`get_sources`) - "Which of our pages rank on Google but AI never cites?" (`get_search_console`) - "Does AI describe our product accurately on every engine?" (`get_brand`) ## Fix it in the codebase This is where a coding agent earns its keep: GetIntel says what is missing, and the agent edits the real site. - "Take the top on-page task, read its brief, and make the change in this repo. Open a pull request when it's done." (`list_tasks`, `get_task`) - "For the comparison topic, check the content analysis and update our comparison page so it covers every fact AI compares." (`get_topic`) - "Look at the latest technical scan and fix the issues on our key pages." (`get_technical`) ## Manage the plan These need a token with action access. - "Plan my week." (`plan_week`) - "Mark the Reddit task as in progress and add it to this week's plan." (`update_task`) - "I've fixed the robots.txt issue. Re-run the technical scan." (`recheck_technical`)