About Loop Factory MCP Public
Loop Factory MCP Public is an MCP server published by alexalexalex222 in the Communication category: loop Factory MCP — a local-first supervisor for evidence-gated AI agent loops. Mines past agent sessions for workflows that worked, improves them generation over generation, and only promotes a measured, re-verified win. Zero dependencies, MCP over stdio.
Recent commit activity is the strongest signal that reported issues will get attention.
Install
git clone https://github.com/alexalexalex222/Loop-Factory-mcp-publicThis server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.
Tools (25)
artifact_recordpersist a raw artifact + sha256; `role:"baseline"` hash-locks (write-once); `measurement` makes the MCP **derive** a tool-computed measurement from the bytes; pass explicit `content` (`sourcePath` reads refused)
Parameters not published.
benchmark_freeze_maker**bench-maker only** — freeze a scorecard directly with `benchSource:"maker"` (bypasses worker `benchmark_propose`); defaults `benchPartition:"gate"` for held-out eval
Parameters not published.
benchmark_runset the tool-**computed** baseline bar; a caller-reported measurement is rejected
Parameters not published.
campaign_statusread-only lane/target queue, auto-transitions, 30-batch retirement + 10–15 advisory accounting, **active `modelPolicy`**, pending dashboard review (never blocks)
Parameters not published.
continue_runrecords the next lane + first concrete action; it does **not** clear the obligation until a real progress tool runs
Parameters not published.
cycle_decision_request**the supervisor hook** — a worker proposes a transition packet (promote/advance_phase/change_baseline/change_benchmark/saturate); only a supervisor-accepted transition is progress; completion/stop intents refused
Parameters not published.
execute_full_test**opt-in (`SUPER_LOOP_ALLOW_EXEC=1`)** — the supervisor itself launches 3–5 allowlisted workers (native executables use direct shell-free `execFile`; allowlisted Windows `.cmd`/`.bat` shims use a narrow `cmd.exe` adapter; prompt always travels via stdin), captures output, parses real token usage, and gates on the tool-captured bytes; off by default → `EXEC_DISABLED`
Parameters not published.
export_trajectoriesread-only export of recorded tool actions as Hermes JSONL with supervisor verdict labels; **refuses gate-partition runs**
Parameters not published.
host_capability_preflightlocal report of which frontier-agent CLIs are installed on PATH (filesystem stat only, never executes, not SOTA/web research) **plus the resolved host profile** — `driverFamily`, `tier`, `setupHint`, and the full host matrix when `SUPER_LOOP_HOST` is unknown
Parameters not published.
host_runtime_detectadvisory guess of which host runtime the agent is in, from which MCP config files exist on disk (per the host registry); read-only existence check — never reads file contents or mutates config; `SUPER_LOOP_HOST` is authoritative when set
Parameters not published.
Show all 25 tools
human_review_requestqueue/list Approve/Sludge items only; model-callable resolve is blocked
Parameters not published.
initialize_loop_runask-once (brief + a few short Qs: goal, **path picker** (improve / discover / mine + library scout), the loop/domain, **corpus scope + order**, what "better" means, a hard limit, **which models** (enter = defaults, `any model` = banlist off), deeper-explanation; promotion mode / standing guarantees stay tool-owned); persists `state.config.modelPolicy`; stores every user message with a sha256 hash; surfaces the stop-condition notice, the **cold-start notice** (fresh run), and the **native-continuation notice** (Claude/Codex `/goal`; `/loop` = Claude's polling alternate) up front; returns a **host-aware `hostSetup`** with a path-aware step 3; honors the "deeper explanation" answer in the same response
Parameters not published.
loop_librarylist mandated (hash-locked) + custom local loops
Parameters not published.
loop_register**add your own loop** to the local MCP: hash-lock, safe id, sectionize, persist locally; never overwrites a mandated loop
Parameters not published.
loop_startbegin phase-gated streaming of any loop (mandated or custom); returns section 0 only
Parameters not published.
observation_recordlightweight phase evidence
Parameters not published.
promotion_requestpromote only on measured + reverified frontier movement; a quality win the MCP can't tool-verify routes to the dashboard (`QUALITY_UNVERIFIED`)
Parameters not published.
register_hypothesesStandard mode: 3–5 hypotheses. Strict real-test mode: exactly two substantive, supervisor-ID-bound hypotheses for one finding. Benchmark-first; rejects banned routes and shape-only placeholders.
Parameters not published.
report_exportreproducible markdown campaign report
Parameters not published.
report_saturationmark a lane saturated → supervisor **auto-transitions** to the next lane (Strip Miner → Loop-de-loop); never pauses/stops
Parameters not published.
reverify_run**re-derive** metrics from the sealed raw bytes and confirm they reproduce (a tampered number cannot survive)
Parameters not published.
run_campaign**autonomous supervisor (opt-in `SUPER_LOOP_ALLOW_EXEC=1`)** — drives the mine→improve queue, validates every worker, measures/reverifies challengers, and keeps review nonblocking. Each changed work epoch gets at most one empty mining pass: empty or duplicate output returns `idle-no-new-work (NOT completion)` instead of repeating paid calls. The standalone CLI stays alive in zero-inference idle and resumes from its target inbox. `maxBatches` is a safety cap, not completion.
Parameters not published.
skill_fetchretrieve skill knowledge for the current task — `plan` mode returns an index of matching skills (titles, purposes, token estimates) to pick from; `section` mode fetches one section body by (`skill_id`, `section_id`); default partition `working`, `reference` is opt-in/held-out only
Parameters not published.
test_hypothesisone full test = 3–5 frontier agents, each tool-computed; aggregates vs the bar; reports quality authority
Parameters not published.
update_dashboardrender the polished always-on local dashboard with the stop-condition notice
Parameters not published.
Ask AI
Ask AI about Loop Factory MCP Public
Powered by Claude · Grounded in docs
Security checks
6 checks have not been run yet.