A Claude loop is a trigger → agent → tools → output cycle that runs on a schedule, event, or command. ClaudeLoops gives you the guides, the working code, and the one-line deploy for every loop worth building.
Plug Claude Desktop, Claude Code, and Projects into one secure URL. Lockers, secrets, and workflow graphs — without shipping API keys to the model.
{
"mcpServers": {
"claudeloops": {
"url": "https://claudelocker.lovable.app/api/mcp",
"transport": "streamable-http",
"headers": { "Authorization": "Bearer lk_demo_<your-token>" }
}
}
}Master these four and you can build any loop on the site. Every guide is structured around them, so shipping the fifth loop feels like shipping the first.
Cron, webhook, Slack command, GitHub event, or a manual /run. When should the loop wake up?
A Claude prompt with a clear role, memory, and tools. Skinny system prompt, generous examples.
MCP servers, HTTP calls, database queries. Each tool is one verb Claude can invoke, no more.
Slack, PR, email, dashboard, ticket. The loop is only real when someone can see or use the result.
Wake up to a 5-bullet diff of every competitor page that changed overnight.
HMAC-verified webhook → Claude enrichment → HubSpot upsert with score.
Claude Code reviews every PR and posts a risk-banded comment with suggested tests.
Stripe + Postgres metrics → narrative artifact → Resend to your investor list.
Drop an .m4a, get a spec doc and a Linear issue with acceptance criteria.
Notion → pgvector → Claude with strict citation schema.
Ten categories. 222 loops. Each with the exact secrets, MCP servers, prompt, code, and deploy command.
Loops that ride inside your repo — PR reviewers, test-gap finders, changelog writers.
Claude Code reviews every PR and posts a risk-banded comment with suggested tests.
Claude Code writes Vitest + Playwright tests for every file touched in a PR.
Claude Code reads schema.prisma diffs, writes safe SQL with rollback notes.
New issue → labeled, assigned, and linked to similar issues in 8 seconds.
When code changes, the matching docs page changes in the same PR.
Auto-rewrites unhelpful PR titles into conventional commits.
Every loop on ClaudeLoops installs the same way. Clone the loop, bind your secrets, deploy. The CLI wires the MCP server, schedules the trigger, and streams logs back.
Every loop opens with the problem, the outcome, and a step list — the code is the last thing, not the first.
The CLI binds env vars from your locker. Claude and MCP tools see references, never raw values.
Every loop run writes a trace: inputs, tool calls, outputs. Replay any run, diff any change.
Cron, GitHub events, Slack slash commands, and Stripe webhooks are one config line, not a side project.
Skip the blank prompt and the glue code. Pick a Claude loop, follow the guide, ship it in an afternoon.