Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Beschreibung vom Entwickler (englisches Original)
Reads your own Ableton projects and writes MIDI, chains and markers into a running Live set.
Diese drei werden nie zu einer Punktzahl zusammengefasst: Meinungen, Messungen und Beliebtheit sind verschiedene Dinge.
Für diesen Server haben wir noch keine Zusammenfassung in einfacher Sprache geschrieben. Unten stehen die Fakten, die wir aus der Registry gelesen haben.
Wähle die KI-App, die du nutzt, und kopiere die Einstellungen.
Welche App nutzt du?
Wähle oben deine App. Wir merken sie uns, damit der Button beim nächsten Mal „In … installieren“ anzeigt.
Die .mcpb-Datei des Entwicklers herunterladen
Der Link führt zur Datei des Entwicklers selbst; wir hosten sie nicht und verpacken sie nicht neu. Öffne die Datei, dann installiert Claude Desktop sie.
Führe das in deinem Terminal aus
Dein Browser fragt, ob er die App öffnen darf, dann zeigt die App die Einstellungen und bittet dich um Bestätigung. Schlüsselwerte bleiben leer, damit du sie selbst einträgst.
Dein Browser fragt, ob er die App öffnen darf, dann zeigt die App die Einstellungen und bittet dich um Bestätigung. Schlüsselwerte bleiben leer, damit du sie selbst einträgst.
Dieser Server läuft auf deinem Computer, und ChatGPT kann sich nur mit Servern im Internet verbinden. Wähle eine andere App.
Deine Auswahl bleibt in einem Cookie auf diesem Gerät; kein Konto nötig.
Diese Einstellungen werden aus dem Registry-Eintrag des Entwicklers erzeugt. Wenn in der README etwas anderes steht, folge der README.
Läuft auf deinem Computer; die Tool-Liste erscheint nach der Installation in deiner KI-App.
Wir prüfen Informationen, die jeder sehen kann; den Code haben wir nicht gelesen. Checks, die nicht zutreffen, werden nicht mitgezählt. · Geprüft am 8. Nov. 2026
Der Registry-Name wurde mit dem GitHub-Konto senolsahan037-oss bestätigt, dem das Repo gehört
Dieses Repo haben wir noch nicht von GitHub gelesen
Dieses Repo haben wir noch nicht von GitHub gelesen
Dieser Server läuft auf deinem Computer; es gibt keinen Endpunkt im Internet
Läuft auf deinem Computer und nutzt stattdessen lokale Einstellungen
Läuft auf deinem Computer; Tool-Listen können wir nur von Endpunkten im Internet lesen
Verglichen mit allen aktiven Servern dieser Kategorie in der Registry. Sterne zählen nur, wenn dem Herausgeber das Repo gehört.
Er hat 1 von 1 zutreffenden Basischecks bestanden. Den Code haben wir nicht gelesen. Bevor du einen Schlüssel oder Token hinzufügst, sieh nach, welche Zugriffe der Server verlangt, und nutze einen Schlüssel, der nur die Berechtigungen hat, die du brauchst.
Die Registry nennt keinen erforderlichen Schlüssel, manche Server bitten dich aber bei der ersten Nutzung, dich anzumelden.
Er läuft auf deinem Computer. Deine KI-App startet ihn bei Bedarf, und die Daten laufen nicht über die Server des Entwicklers (außer der Server ruft selbst einen Onlinedienst auf).
Das GitHub-Repo wurde zuletzt – geändert, und die neueste Version (0.4.4) kam vor 2 Monaten in die Registry.
Apps, die MCP unterstützen, etwa Claude Desktop, Claude Code, Cursor und VS Code. Außerdem brauchst du MCPB auf deinem Computer.
In derselben Kategorie: Spec Driven Development, Remote Desktop Commander, Compose Preview Catalogs
Server derselben Kategorie von anderen Entwicklern, sortiert nach Sternen.
So liest du eine Karte:↓ Downloads pro Wochei★ GitHub-SterneiDrei Säulen: Bewertungen · Checks · NutzungiAktualisiert = neueste Version in der Registry veröffentlichti
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Browse, inspect and render Jetpack Compose Material 3 and Wear component catalogs.
Unverändert aus dem Repo des Entwicklers übernommen. PopNx MCP hat den Inhalt weder geprüft noch empfiehlt es ihn. HTML, Bilder und Skripte wurden entfernt. Links führen von dieser Website weg.
Loom is a measurement-based production system for Ableton Live by Şenol Şahan / SubverseLab: a local MCP server that reads your own .als projects and library, answers with counts instead of guesses, and writes MIDI, device chains, automation and arrangement markers into a running Live session, verifying every write by reading it back.
Canonical home: https://subverselab.com/loom · Cite: CITATION.cff · Licence: AGPL-3.0-or-later · Attribution: NOTICE.
Ask a general-purpose model what belongs on a kick drum and it answers with confidence, having never heard a record you made. Loom replaces that guess with evidence from three places: your own project archive (which devices you actually chain, which samples you actually reach for), real released records (genre evidence measured from audio, not described from memory), and the Live set that is open right now. Where the evidence runs out, a tool returns nothing and says so. Nothing is emulated, nothing is retried on an uncertain outcome, and every write is a dry run until you approve it.
There is one connection to Live: the Loom extension, an Ableton Extension running inside Live 12.4 beta. No Control Surface, no automatic fallback, no second writer path.
| Layer | Job |
|---|---|
| Sensei | MIDI variation (drum / bass / chord) from a locked, hash-verified dataset |
| ArrangementGPS | A project plan from a prompt: tempo, key, genre, channels, sections (Node) |
| AIMixMaster | .als inspection, gain staging, clip alignment, drum buss, automation writing (on the file) |
| Presetor / AISoundDesigner | Device chains and a sound palette measured from the user's own projects |
| MusicalIntelligence | Genre evidence measured from real records; part suggestions keyed to the project |
| Mix Check / SampleAgent | Audio measurement against profiles of released masters; YouTube → sliced sample pack; a crate-digging agent |
| Loom extension | The Ableton Extension inside Live: the MCP's only connection to a running set |
| mcp_server | 45 tools, resources, prompts, progress and cancellation over stdio JSON-RPC |
No layer produces a guess: with no evidence there is no suggestion, what the SDK cannot do is not emulated, and every writing tool is a dry run by default.
Architecture, the canonical call flow and the bridge protocol: Docs/ARCHITECTURE.md.
python3 install.py # installs
python3 install.py --check # changes nothing, reports state
install.py:
extension/dist/loom.ablx, built if the toolchain is present) and compares the version and bridge protocol of the extension installed in Live with this checkout. Live shows it as Loom, id subverselab.loom. An installed legacy package (loom.sensei-midi-writer) is reported: the MCP never sends it a mutation (LEGACY_EXTENSION).Live's own step is a single one: add the .ablx in Live 12.4 beta's Extensions settings and restart Live. The extension opens a file bridge in its own storage directory; the MCP locates that bridge on every call.
The same server is packaged as an MCP bundle on the release page (loom- .mcpb, described by server.json).
Versions are three different things: the package version is written once, in extension/manifest.json (package.json must match; the build verifies it); the bridge protocol is loom.bridge/3; the SDK API is 1.0.0.
The MCP tool `live_bridge_status` reports which bridge it is talking to and why, the age of the state, the session id, the capabilities the extension publishes, mutations_allowed, the protocol verdict (OK / UPGRADE_REQUIRED / PROTOCOL_MISMATCH / STALE_STATE / NO_STATE), the journal's condition and what is queued or in flight. python3 install.py --check prints the same in a terminal.
An old extension (0.1.0 / 0.2.0, publishing no protocol) is read but never written: every mutation is refused with UPGRADE_REQUIRED before a request file is written. A package installed under the legacy id is treated the same (LEGACY_EXTENSION); once the new one is installed and the old one removed from Live, the old journal is carried over with live_command op=journal_import (nothing forgotten, nothing deleted). More than one Loom bridge in sight is AMBIGUOUS_BRIDGE. LOOM_BRIDGE_ROOT exists for tests only.
| Kind | Tools | Proven by |
|---|---|---|
| Through Live (extension) | live_state, live_bridge_status, live_command (set_tempo, set_mixer, set_device_parameter, list_device_parameters, create_locator, create_midi_track, import_audio_clip, render_pre_fx, drum_pads, build_drum_kit, journal_import, delete_track, delete_locator), midi_write_arrangement, midi_write_to_live (session clip), crate_to_live, mix_from_live, project_build, midi_generate (auto_write) | mcp_server/tests/test_bridge_consumer_real.py (the real bridge.ts), test_extension_path.py, extension/tests/bridge.test.ts |
| Engines without Live | project_*, automation_*, drumbuss_*, chain_*, render_*, palette_read, library_search, genre_evidence, part_suggest, plan_create, plan_verify, projects_arrangement_shapes, mix_measure/analyze/profiles, crate_fetch/read/spots/chop/agent, setup_scan, gap_record | mcp_server/tests/test_mcp_tools.py (45 tools over stdio) and each engine's own pytest suite |
| OS level, on request | live_project (open / close / status of Live, verified from Live's own log; changing the set restarts the extension host) | test_live_project.py |
| Machine-bound | mix_capture (method="tap": a Core Audio process tap launched as LiveTap.app through LaunchServices; macOS grants the "Screen & System Audio Recording" permission to the LiveTap entry, not to the app running the MCP; measured from a playing Live on 2026-09-06) | only on a real machine |
The extension's single in-Live user command, "Loom: Generate" (right-click, Session slot), goes through the same write_clip implementation and the same ownership ledger: a clip Loom did not write is never overwritten.
The SDK does not load presets (.adg / .adv). Loom does not call this "impossible"; it separates the paths:
project_build(kit="Boom Bap Kit") or a path to an .adg. The kit is read from the preset's own XML (pads, notes, names, sample files; Sensei's .adg reader), the files are resolved on this machine, and the pads are rebuilt in the extension as chain + Simpler + sample. The answer states what was carried over and what was not: per-pad effects, macros, choke groups, Simpler parameters, return chains. This is not loading the preset as is, and it is never presented as such.live_command op=build_drum_kit pads=[{note, sample}].device_map (Operator, Electric, Wavetable, …), or you load the preset in Live yourself and re-run the same plan; the channel is adopted and its device read from state. The answer lists both routes under needs_preset.pad_mapping: by_role), unmappable roles are dropped and reported, and with no notes left no clip is written (no_notes_for_pads).tracks=[...] and device_map the answer's simplification block says which channels were dropped and why, and which device stands in for which preset. Empty channels from the template are left alone..als (automation_write, drumbuss_build, chain_apply) change the set on disk; the open set is not touched and must be reopened. Preset XML is never hand-converted into a set; that is Live's job.Extensions SDK 1.0.0-beta.1 does not provide the following. Loom does not emulate them; it answers UNSUPPORTED_BY_SDK with the name of the missing capability, before any request file is written:
live_command op=transport, mix_capture follow_transportlive_command op=set_key; the project_build step is reported as UNSUPPORTED_BY_SDKcreate_midi_track only adds a native device with its default preset (not_loadable_in_extension); kits are rebuilt (above), instruments go through device_map or a user stepbeats_per_bar, reads it from the .als when given one, and otherwise says it assumed 4/4 in beats_per_bar_sourcemix_capture method="resample", the capture_* opsReported to Ableton through Centercode on 2026-09-03.
Every answer from Live carries a structured outcome: {kind: applied|refused|failed|indeterminate, code, applied, verified, side_effects, next_step}. The MCP status derives from it: OK, REFUSED_IN_LIVE (Live untouched), FAILED_IN_LIVE (attempted, previous content restored), INDETERMINATE (may have applied; side effects and the safe next step are in the answer), NOT_CONSUMED, INVALID_RESULT. An uncertain outcome is never retried automatically, anywhere.
Keyed requests (build steps, idempotency_key) are journaled before the mutation: same key + same content → the stored result; different content → a conflict; another session → refused; half-finished → INDETERMINATE. With the journal missing or corrupt, keyed mutations are refused and the answer says what to do; deleting the journal is never the advice. Details in Docs/ARCHITECTURE.md.
./scripts/check_ci.sh # needs neither Ableton nor personal data; CI runs this
./scripts/check_all.sh # everything; needs a real Ableton install
check_ci.sh counts every suite as passed, failed or skipped; a suite with a missing dependency is never counted as passed. The extension's own queue code (bridge.ts) runs both on its own (npm run test:bridge) and as the real consumer opposite the MCP (test_bridge_consumer_real.py); the Python fake exists for speed and is not protocol proof on its own.
Tests use temporary directories and isolated fixtures; they never touch a real Live, the installed extension or user data. The Ableton Extensions SDK is not redistributable, so the hosted runner cannot typecheck the extension; that job is marked skipped, never passed.
The real-Live acceptance script is extension/tools/measure_bridge.py (it writes into the running Live's open set; run it on an empty set).
This repository ships code and fixtures, never measurements. Presetor's device-chain evidence, AISoundDesigner's palette and Sensei's catalogs are generated from the user's own projects and Ableton installation; none o
Nur der Anfang wird gezeigt; den Rest findest du im Repo