For agents

Use the looprepo catalog directly — no install required. Every loop is graded A–F by a default-fail evaluator, so an agent can pick a safe loop without a human.

Drop-in prompt

Paste this into any coding agent to let it query the live catalog on its own — it works verbatim, no edits needed.

You have access to looprepo, a directory of copy-paste AI agent loops. Do not rely on memory or the paginated site. Read fresh data: 1. Fetch https://looprepo.dev/catalog.json for the structured catalog. 2. Filter by category, loopType, agents, or tags. 3. Prefer loops with a safety grade of A or B. 4. Read the loop's url for the full prompt and stopping conditions. 5. Copy the prompt verbatim; swap in the user's real repo and constraints. Never invent missing project details.
→ Claude

Endpoints

/catalog.json

Structured source of truth for programmatic selection. Includes safety grade, loop type, cost/risk, difficulty, agents, and tags per loop.

/catalog.txt

Raw plain-text catalog for agents without JSON tooling.

/llms.txt

Concise operating instructions and safety boundaries.

https://mcp.looprepo.dev/mcp

MCP endpoint for tool-based access — search, get, categories, grade, and harness generation. Install below.

MCP server

Tool-based access over Streamable HTTP at https://mcp.looprepo.dev/mcp full docs. No auth, no API key — read tools are open; rate limit 60 req/min per IP. Claude Code:

claude mcp add --transport http looprepo https://mcp.looprepo.dev/mcp
→ Claude

Cursor / Windsurf / VS Code — add to your MCP config:

{ "mcpServers": { "looprepo": { "url": "https://mcp.looprepo.dev/mcp" } } }
→ Claude

22 tools across four groups — read tools are free, the two LLM cost centers (crafting loops, building harnesses) are daily-capped. Full docs and limits →

ToolWhat it doesCost
Discovery
loops_searchsearch the catalog by text, category, loop type, or safety gradefree
loops_getfetch one loop’s full prompt and stopping conditions by slugfree
loops_categorieslist categories and loop types with live countsfree
loops_trendingnewsurface loops gaining copies and upvotes right nowfree
loops_similarnewfind loops close to a given slug by category, type, and tagsfree
loops_for_stacknewmatch loops to a described stack — framework, tools, agentsfree
loops_comparenewput two or more loops side by side on safety, scope, and costfree
loops_statsnewdirectory totals — counts by grade, category, and loop typefree
loops_of_the_weeknewthe current Loop of the Week — top-voted, refreshed weeklyfree
Grading & safety
loops_gradegrade any loop prompt A–F with the deterministic, default-fail gatefree
loops_hardennewrewrite a loop to add missing exit conditions, budgets, and gatesLLM · capped
loops_grade_explainnewplain-language breakdown of why a loop earned its gradeLLM · capped
harness_validatenewcheck a harness bundle for the required safety scaffoldingfree
Build & intake
loops_generate_harnessgenerate a complete safe-loop harness bundle for a goalLLM · capped
loops_intake_recommendturn a read-only Claude-use report into ranked loop matches plus a builder prefillfree
loops_craftnewdraft a net-new candidate loop from a described workflow, staged through the same gateLLM · capped
intake_prompt_getnewreturn the read-only Claude-use intake prompt to paste into an agentfree
loops_submitnewsubmit a loop to the review queue — text only, sanitized, gatedfree
Account
cme_tier_statusnewshow separate Looprepo rolling allowances and CME plan entitlements; the optional verified account email adds its build counterfree
beacon_run_reportnewreport a loop run to Mission Control telemetryfree
articles_searchnewsearch looprepo’s published loop-engineering articlesfree
articles_getnewfetch one article’s full text by slugfree

Why the Safety Score matters here

Loops run autonomous agents. A grade is not decoration — it's the guardrail.

Each loop carries an exit condition, an iteration budget, a verification check, and tight scoping — and a grade that reflects them. Agents should prefer A/B loops and keep those guards in place. How review works →