looprepo
The destination for AI agent loops. Copy-paste /loop, /goal, /schedule and Ralph-style prompts for Claude Code and Codex, or generate a full safe-loop harness.
https://looprepo.dev/Connect straight to this server’s public endpoint.
https://looprepo.dev/api/mcpWe add this server to your workspace, walk through sign-in once, then open Studio with tools ready to run.
Last probed Sep 14, 2026 · looprepo.dev
22tools discovered
Search agent loops
Search the looprepo directory of agent loops (/loop, /goal, /schedule, Ralph-style) for Claude Code and Codex. Returns matching loops with their copy-paste prompts, ranked by upvotes and copies.
Grade a loop prompt
Grade any agent-loop prompt A–F with the same deterministic, default-fail Safety Score that gates the looprepo directory. Returns grade, score (0–100), gate verdict (pass/fail), and per-signal reasons (exit condition, turn cap, verification, blast radius). Purely deterministic — no LLM. For an AI explanation of a grade, use looprepo.dev/grade.
Get loop by slug
Get one loop by slug: full prompt, usage notes (when to use / how it works / safety), attribution, and stats.
List categories & loop types
List all categories and loop types with published-loop counts. Useful before searching.
Recommend loops from a Claude-use intake report
Given a read-only intake report pasted from Claude, Claude Code, Cowork, Codex, or another agent, infer the goal/workflows and return ranked real looprepo loops plus a harness-ready builder prefill. Works for code projects and general Claude use cases. Does not invent loop titles, run code, or require repo access.
Generate safe-loop harness
Generate a complete safe-loop harness bundle. Claude Code target: CLAUDE.md, .claude/settings.json PreToolUse+Stop hooks, slash commands, fresh-context default-fail evaluator agent. Codex target: AGENTS.md, read-only evaluator.md, and a sandboxed codex-exec run-loop.sh that enforces the iteration budget. Both include prd.json and progress/handoff templates. Free (soft anti-abuse cap: 30/day per email).
Trending loops
Top loops by upvotes + copies over a recent window. Optional category/loop_type filters.
Similar loops
Loops related to a given slug — same category, matching intent, and text similarity.
Loops for your stack
Describe your tools/stack (e.g. "next.js railway playwright postgres") and get ranked matching loops.
Compare loops
Compare 2–3 loops side by side: safety grade/score, type, risk, and boundedness signals.
Directory stats
Directory-wide stats: totals, category/type breakdown, safety-grade distribution, newest loops.
Loop of the week
The top loop of the trailing 7 days plus recent A/B-grade picks.
Search guides
Search looprepo's published guides/articles on loop engineering.
Get guide by slug
Fetch one published guide: full markdown body plus metadata.
Submit a loop for review
Submit a community loop to the review queue. Submissions are text-only, sanitized, and NEVER auto-published — an owner reviews every one against the default-fail safety gate before it can appear in the directory.
Get the Claude intake prompt
Returns the read-only Claude-use intake prompt. Paste it into any Claude context, then send the resulting report to loops_intake_recommend.
Validate a harness bundle
Deterministically validates the persistent harness separately from the loop running inside it, then diagnoses likely failure modes. Checks standing instructions, skill routing, permissions/hooks, state, on-disk goal, bounded action, independent verification, iteration budget, stop condition, and default-fail behavior. Pure text analysis — nothing is executed.
Harden a loop prompt
Deterministic grade plus the exact edits to reach the next grade (exit condition, turn cap, verification, scope). Same scorer that gates the directory — no LLM.
Explain a safety grade
Deterministic A–F grade plus an AI-written explanation and top-3 edits. The grade is never LLM-derived; the explanation is advisory and falls back to deterministic advice when unavailable. Optional cme_api_key raises the daily AI limit for CME Pro/Power members.
Craft a loop from a description
Describe a workflow in plain language and get one safe, copy-paste loop prompt with verification, stop conditions, and approval gates. Draft-only. Optional cme_api_key raises the daily limit for CME Pro/Power members.
Check your tier & limits
Your current tier (anonymous or CME Pro/Power), daily limits, and today's usage of crafted loops and harness builds.
Mission Control run report
Recent loop runs for your Mission Control beacon token: status, exit reason, turns, cost.
Get your MCP into directories
A working endpoint is step one. Directory coverage is the coordinated launch across ChatGPT, Claude, Cursor, the MCP Registry, and community indexes.
Directory coverage for brandsThe destination for AI agent loops. Copy-paste /loop, /goal, /schedule and Ralph-style prompts for Claude Code and Codex, or generate a full safe-loop harness.
Use the MCP endpoint listed on this page in your MCP client configuration. One-click install pills support Claude, Cursor, VS Code, and other hosts. Copy the remote MCP URL if your client needs a manual entry.
MCPBundles probed 22 tools on the live server. The tool list on this page reflects what was discovered at the last refresh — connect your client to see the full set available to your session.
looprepo may require signing in to the provider before tools can run. Connect through MCPBundles or your MCP client and complete any provider login when prompted.
Operate looprepo? Verify ownership to take over this directory entry.
This server appears in the MCPBundles directory. Verify you operate it to take over the listing — name, description, logo, contact email, and skill content. We email a 6-digit code to a maintainer address your server publishes in /.well-known/security.txt or /.well-known/mcpbundles.json. Free, takes about a minute.