Grok Bot + Claude Code: hand X research to Grok mid-task
Claude Code calls a Grok Bot routine through a webhook whenever it needs current X posts, waits for the answer, then finishes the job, such as a landing page or a script.
Setup
- In Grok Bot, create a routine that starts from a webhook. It gets its own webhook URL and key, and the request body reaches the Bot.
- Write a Claude Code skill, for example .claude/skills/grokbot-research/SKILL.md, that sends the question to the webhook and waits for the reply.
- Decide how the answer gets back. The demo's guide covers this behind a sign-in; a simple option is for the routine to save its answer to a doc or GitHub issue the skill reads.
- Test with one question, and check the cited posts before you build on them.
Starter prompt
---
name: grokbot-research
description: Use when the task needs current posts or discussion from X.
---
1. Send the question to the Grok Bot routine's webhook URL, with its webhook key, asking for: the 10 most relevant X posts with links, and a 5-line summary.
2. Wait for the answer at [doc / issue the routine writes to]. If nothing arrives within 10 minutes, carry on without it and say so.
3. Cite the post links you use. Never invent posts or quotes.
Business fit
How this stack helps specific businesses:
Watch-outs
- The demo comes from Agent Native, which publishes the setup guide; no business results have been published yet.
- Several replies argue the reverse works better: Grok Bot coordinating and Claude Code doing the building. See the next stack.
- Treat the webhook URL and key like passwords.