Context7
Up-to-date docs for AI agents
Beschreibung vom Entwickler (englisches Original)
Persistent memory for AI agents — shared .brain across Claude Code, Cursor & Codex
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.
NUCLEUS_BRAIN_PATHDu brauchst
Welche App nutzt du?
Wähle oben deine App. Wir merken sie uns, damit der Button beim nächsten Mal „In … installieren“ anzeigt.
Funktioniert in Claude im Web, auf dem Desktop und auf dem Handy:
https://relay.nucleusos.dev/mcpÖffne Claude Desktop → Settings → Developer → Edit Config und füge dies in claude_desktop_config.json ein
{
"mcpServers": {
"nucleus": {
"command": "uvx",
"args": [
"nucleus-mcp"
],
"env": {
"NUCLEUS_BRAIN_PATH": "your-value"
}
}
}
}Führe das in deinem Terminal aus
claude mcp add --transport http nucleus https://relay.nucleusos.dev/mcpclaude mcp add nucleus --env NUCLEUS_BRAIN_PATH=your-value -- uvx nucleus-mcpTrag die leeren Werte ein, bevor du ihn startest.
In .cursor/mcp.json in deinem Projekt oder in ~/.cursor/mcp.json einfügen
{
"mcpServers": {
"nucleus": {
"url": "https://relay.nucleusos.dev/mcp"
}
}
}In .cursor/mcp.json in deinem Projekt oder in ~/.cursor/mcp.json einfügen
{
"mcpServers": {
"nucleus": {
"command": "uvx",
"args": [
"nucleus-mcp"
],
"env": {
"NUCLEUS_BRAIN_PATH": "your-value"
}
}
}
}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.
In .vscode/mcp.json in deinem Projekt einfügen
{
"servers": {
"nucleus": {
"type": "http",
"url": "https://relay.nucleusos.dev/mcp"
}
}
}In .vscode/mcp.json in deinem Projekt einfügen
{
"servers": {
"nucleus": {
"type": "stdio",
"command": "uvx",
"args": [
"nucleus-mcp"
],
"env": {
"NUCLEUS_BRAIN_PATH": "your-value"
}
}
}
}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.
ChatGPT verbindet sich nur mit Servern im Internet:
https://relay.nucleusos.dev/mcpMenünamen können sich je nach Tarif und mit der Zeit leicht unterscheiden.
Deine Auswahl bleibt in einem Cookie auf diesem Gerät; kein Konto nötig.
NUCLEUS_BRAIN_PATH envDiese Einstellungen werden aus dem Registry-Eintrag des Entwicklers erzeugt. Wenn in der README etwas anderes steht, folge der README.
Tool-Namen und -Beschreibungen genau so, wie der Server sie am 6. Nov. 2026 gemeldet hat. Geschrieben hat sie der Entwickler, nicht wir; wir zeigen sie als reinen Text.
nucleus_governanceGovernance, Hypervisor & security tools for the Nucleus Agent OS. Actions: auto_fix_loop - Auto-fix loop: Verify->Diagnose->Fix->Verify (3 retries). params: {file_path, verification_command} lock - [HYPERVISOR] Lock a file/dir immutable (chflags uchg). params: {path} unlock - [HY
nucleus_featuresFeature tracking, proof generation & MCP server mounting. Actions: add - Add a feature. params: {product, name, description, source, version, how_to_test, expected_result, status?, tags?} list - List features. params: {product?, status?, tag?} get - Get feature by ID. params: {fe
nucleus_sessionsSession management, events, state & checkpoint tools. Actions: save - Save session for later. params: {context, active_task?, pending_decisions?, breadcrumbs?, next_steps?} resume - Resume a saved session. params: {session_id?} list - List all saved sessions check_recent - Check
nucleus_tasksTask management, depth tracking & ADHD context-switch tools. Actions: list - List tasks. params: {status?, priority?, skill?, claimed_by?, required_role?} get_next - Get highest-priority unblocked task. params: {skills, required_role?} claim - Atomically claim a task. params: {ta
nucleus_syncSync, artifact, trigger & deploy management for multi-agent coordination. Actions: identify_agent - Register agent identity. params: {role, provider, session_id} (per ADR-0005 §D1) OR legacy {agent_id, environment, role?} (coerced per §D5 until end of Cycle C+2) sync_status - Che
nucleus_next_messageGet the next unread relay message. One call — returns full message body and auto-acks. Long-poll (60s timeout). This is the agent's idle loop. Call it, get a message or None, act or re-call. No separate inbox + ack dance. The agent's loop: 1. nucleus_next_message() → blocks up to
nucleus_relay_subscribeLong-poll subscription that pushes ctx.info() on each new inbox file. Replaces bash polling daemons (watch-relay-*.sh) with server-initiated push. Call once at session start (e.g. via SessionStart hook). Server holds the subscription, watches the calling agent's role-specific inb
nucleus_ccr_armOne-shot convenience: resolve canonical inbox + arm long-poll subscription. Per PR #2 (CCR server-side auto-arm) — IDE-agnostic relay-arrival arming. This is the RECOMMENDED entry point for SessionStart auto-arming across all MCP clients (Claude Code, Antigravity, Cursor, Windsur
nucleus_wakeup_waitQuick scan for a PENDING task. Returns the task directly if one is available, or None if no task is ready within the timeout. Default timeout is 5s (non-blocking). The agent should NOT loop on this — tasks arrive via relay push. This is a fallback for when the agent wants to chec
nucleus_orchestrationSatellite view, commitments, loops, patterns & metrics. Actions: satellite - Unified satellite view. params: {detail_level?} scan_commitments - Scan artifacts for new commitments archive_stale - Auto-archive commitments older than 30 days export - Export brain to zip list_commitm
nucleus_telemetryLLM tiers, telemetry, PEFS notifications & protocol tools. Actions: set_llm_tier - Set default LLM tier. params: {tier} get_llm_status - Get LLM tier configuration record_interaction - Record user interaction timestamp value_ratio - Get Value Ratio metric check_kill_switch - Chec
nucleus_slotsOrchestration slots, sprints & mission management. Actions: orchestrate - THE GOD COMMAND. params: {slot_id?, model?, alias?, mode?} slot_complete - Mark task complete. params: {slot_id, task_id, outcome?, notes?} slot_exhaust - Mark slot exhausted. params: {slot_id, reset_hours?
nucleus_infraInfrastructure: file changes, cloud, marketing & strategy tools. Actions: file_changes - Get pending file change events gcloud_status - Check GCloud auth status gcloud_services - List Cloud Run services. params: {project?, region?} list_services - List Render.com services scan_ma
nucleus_agentsAgent spawning, critic, swarm, memory, ingestion & dashboard tools. Actions: spawn_agent - Spawn Ephemeral Agent. params: {intent, execute_now?, persona?, confirm?}. HITL: requires confirm=true. apply_critique - Apply critique fixes. params: {review_path} orchestrate_swarm - Star
nucleus_federationFederation management for multi-brain coordination. Actions: status - Get comprehensive federation status join - Join a federation via seed peer. params: {seed_peer} leave - Leave the federation gracefully peers - List all federation peers with details sync - Force immediate sync
nucleus_engramsEngrams, health, observability, DSoR & tier system tools. Actions: health - Get system health status version - Get Nucleus version info export_schema - Export MCP toolset as JSON Schema performance_metrics - Get perf metrics. params: {export_to_file?} prometheus_metrics - Get Pro
nucleus_relayRelay-substrate facade — post / read / ack / status. Actions: post - Send a relay envelope. params: {to, subject, body, sender?, priority?, in_reply_to?, context?, id?, from_session_id?} sender? auto-fills from your session role when omitted. to accepts short role aliases: main,
nucleus_routeW5 tier routing — route a prompt to the optimal model tier. Actions: route - Route a prompt to cheapest capable model. params: {prompt, complexity?, context?, estimated_output_tokens?} complexity ∈ 'routine' | 'complex' | 'sovereign' (default 'routine') Returns: provider, model,
nucleus_auditW8 Team-tier tamper-evident audit log (SHA-256 hash chain). Actions: log_event - Append an audit event to the chain. params: {event_type, actor, resource, outcome, metadata?, team_id?, ts?} query - Single-tenant query. Rejects team_id='*'. params: {team_id, since?, until?, actor?
nucleus_delegateHand a coding or review task to a cross-vendor lane — a cheap, non-Claude agent that does the work for you. Use this INSTEAD OF shelling out to any CLI: it picks the vendor, injects the correct per-vendor permission flags, captures all output synchronously, and reports a real sta
nucleus_lane_initSet up an autonomous task-execution loop in a git repo. WHEN TO USE: you have a list of tasks (bugs, features, tests) you want executed automatically without manual prompting — the lane runs watcher/executor/secretary daemons that claim tasks, invoke an LLM to implement them, and
nucleus_lane_startStart autonomous lane daemons (watcher + executor + secretary) in background. WHEN TO USE: after nucleus_lane_init and SPEC.md is written — this launches the loop that autonomously claims and executes tasks. The executor invokes an LLM CLI (devin/agy) per task; the secretary inde
nucleus_lane_stopStop all running lane daemons. WHEN TO USE: when all tasks are CONFIRMED or you want to pause autonomous execution.
nucleus_lane_statusCheck lane status: task counts (PENDING/IN_PROGRESS/DONE/CONFIRMED), daemon health, spec verification. WHEN TO USE: after starting a lane to monitor progress, or before starting to check if a lane is already running.
nucleus_lane_feedbackReport friction/bugs/enhancements to the nucleus team. WHEN TO USE: when the lane misbehaves, a task fails unexpectedly, the SPEC.md format is confusing, or you have an improvement idea. Creates a GitHub issue on the public nucleus-mcp repo (cross-machine visible) AND stores loca
nucleus_plan_executeExecute a plan file autonomously. WHEN TO USE: you have a plan file (.brain/plans/*.md) you want executed autonomously — parses the plan, creates tasks, starts a mission with budget/time limits. Chains import_plan_as_tasks() then _brain_start_mission_impl() and returns the missio
nucleus_plan_importImport a plan file as PENDING tasks without starting a mission. WHEN TO USE: you have a plan file (.brain/plans/*.md) you want loaded into the task store so the executor daemon / sprint mission can pick the tasks up later — but you don't want to start a mission right now.
nucleus_plan_listList available plan files in .brain/plans with task counts. WHEN TO USE: you want to discover which plan files exist and how many tasks each contains before importing or executing one. Returns: JSON string: a list of dicts each with ``{"name", "path", "size_bytes", "task_count",
Wir prüfen Informationen, die jeder sehen kann; den Code haben wir nicht gelesen. Checks, die nicht zutreffen, werden nicht mitgezählt. · Geprüft am 7. Nov. 2026
Der Registry-Name wurde mit dem GitHub-Konto eidetic-works bestätigt, dem das Repo gehört
Dieses Repo haben wir noch nicht von GitHub gelesen
Dieses Repo haben wir noch nicht von GitHub gelesen
Antwort in 583 ms (HTTP 200)
Keine Anmeldung nötig
Wir haben 28 Tool-Beschreibungen gelesen und keine versteckten Anweisungen an die KI gefunden
Verglichen mit allen aktiven Servern dieser Kategorie in der Registry. Sterne zählen nur, wenn dem Herausgeber das Repo gehört.
Er hat 3 von 3 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.
Beides geht: Installiere ihn auf deinem Computer oder verbinde dich mit der Cloud-Version.
Das GitHub-Repo wurde zuletzt – geändert, und die neueste Version (1.17.1) kam vor 5 Tagen in die Registry.
Apps, die MCP unterstützen, etwa Claude Desktop, Claude Code, Cursor und VS Code. Er verbindet sich per URL, also funktionieren auch Apps, die Remote-MCP-Server akzeptieren, zum Beispiel ChatGPT im Entwicklermodus.
In derselben Kategorie: Context7, Codebase Memory, Exa
Server derselben Kategorie von anderen Entwicklern, sortiert nach Sternen.
So liest du eine Karte:★ GitHub-SterneiDrei Säulen: Bewertungen · Checks · NutzungiAktualisiert = neueste Version in der Registry veröffentlichti
Up-to-date docs for AI agents
Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.
Connect AI agents to Exa for web search, content fetching, and multi-step research.
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.
.brain — the portable decision logThe portable decision log your AI tools all read. One MCP server. Any AI tool. Plain files.
Every AI coding session starts by re-explaining context the last session already knew. .brain is a folder in your repo that Claude Code, Cursor, and Codex all read via one MCP server. Decisions, policies, plans — written once, remembered across every session and every tool.
MIT licensed. File-based (plain JSON + markdown). No embeddings. No vendor lock-in.
nucleus-mcp ships a second, fully independent tool: nucleus-rabbithole, a rabbit-hole depth tracker for focus-prone developers.
It gives your AI a push/pop depth stack, a context-switch thrash detector, an open-loop externaliser, and a weekly review — all backed by local SQLite, no network, no daemon.
# Already installed with nucleus-mcp — just run:
nucleus-rabbithole
Claude Code .mcp.json snippet:
{
"mcpServers": { "nucleus-rabbithole": { "command": "nucleus-rabbithole", "args": [] } } }
Full documentation: docs/RABBITHOLE.md — also installed with the package, at /mcp_server_nucleus/docs/RABBITHOLE.md.
The core loop that makes AI reliability compound over time:
GROUND ALIGN COMPOUND
────── ───── ──────── Machine verifies Human corrects System learns
AI writes code → You fix a mistake → Delta recorded GROUND checks → Verdict stored → DPO pair created Receipt logged → Event emitted → Training data grows │ │ │ └───────────────────┴────────────────────┘ Reliability improves
GROUND — 5-tier execution verification. Syntax, imports, tests, runtime. Goes outside the formal system to check the AI's work.
ALIGN — One-call corrections. nucleus review records a verdict against a prior decision, which becomes a training pair and an event.
COMPOUND — Deltas measure the gap between intent and reality. Recurring patterns become strategy. Negative deltas become training signal.
Every tool response shows frontier health:
[frontiers: GROUND 42 | ALIGN 12 | COMPOUND 28]
Hosted access is not currently available. The remote MCP endpoint is withdrawn pending an access-control review. Use the local install below.
Local install (Cursor, Windsurf, Claude Desktop):
pip install nucleus-mcp
nucleus init --recipe founder
Two commands. Nucleus is running. nucleus init seeds .brain/, writes a project-local .mcp.json, and offers to configure every AI client you have installed. Say yes, then restart the client.
nucleus build additionally needs a coding-agent CLI on PATH — any one of claude (npm i -g @anthropic-ai/claude-code), devin, or agy, already authenticated (claude/devin/agy login once, separately). Without one, build stops at the plan stage. If build reports status='error' without a clear cause, the vendor CLI's own error (e.g. an expired login) lands in .brain/relay/cross_vendor/*.json — check the newest file there.
30 MCP tools across 16 facades, exposing 210 actions, plus 19 brain:// resources and 8 prompts. Regenerate these with python3 tools_audit/count_surface.py rather than editing them by hand — three hand-maintained copies of this number had drifted apart before that script existed.
Benchmark: decision-retention-evals — does your AI agent remember why the code is the way it is? (ask on Discord for the current mirror)
Everything above is free (MIT). Nucleus Pro adds verifiable governance:
nucleus trial # 14-day free trial
nucleus compliance-check # Score your AI governance nucleus audit-report --format html -o report.html # Audit report
$19/month or $149/year — nucleusos.dev/pricing
| Free | Pro | |
|---|---|---|
| 30 tools, 19 resources, 8 prompts | Yes | Yes |
| Persistent memory | Yes | Yes |
| Governance & HITL | Yes | Yes |
| Audit trails (DSoR) | Yes | Yes |
| Signed audit reports | - | not yet shipped |
| Compliance exports | Score only | Full PDF/HTML |
| Priority issues | - | Yes |
pip install nucleus-mcp # or: uvx nucleus-mcp · pipx install nucleus-mcp
nucleus init # seeds .brain/, then offers to configure your AI clients
Requires Python ≥3.10; on 3.14 some dependencies (e.g. watchdog) build from source — 3.11–3.13 install wheels cleanly. init writes whichever nucleus binary it resolves first on PATH into the config it generates — with a stale global install alongside a venv, pipx/uvx or which nucleus keeps the right one.
nucleus init always does three things: seeds .brain/, writes a project-local .mcp.json that Claude Code reads in that folder, and runs a memory write-and-recall self-test so you can see it working before you trust it.
In an interactive terminal it then asks whether to configure your other AI clients. Answer yes and it writes the nucleus entry into every one it finds installed — Claude Desktop, Claude Code, Cursor, Windsurf, Antigravity, Devin — backing up each file it touches. It tells you how many it configured, and says so plainly when that number is zero rather than leaving you to wonder.
Run with --no-wizard, or in a script where there is no terminal to ask, and it seeds the brain and the project-local config only. Use nucleus setup for the client configuration in that case.
Then restart your AI client. To verify: your client's tool list now shows nucleus_* tools, or run nucleus doctor.
An existing nucleus entry is never touched unless you pass --force. For most clients a .json.bak copy is written before the edit. Claude Code's ~/.claude.json is handled differently and deliberately: nucleus writes a temp file and swaps it in atomically, so a failed write leaves the original exactly as it was, and no .bak is left lying beside it. A client that is installed but has never had an MCP config gets one created; a client you do not have is left alone entirely, and ~/.claude.json is never invented if it is missing. Already have a .brain? Run nucleus setup to configure clients without re-seeding it, with --dry-run to preview the exact changes first.
A one-click `nucleus.mcpb` bundle for Claude Desktop is built via bash scripts/build_mcpb.sh (it will be attached to releases once the release workflow ships it). Opening the bundle with Claude Desktop uses Claude's built-in uv runtime to fetch and run nucleus-mcp — no Python setup required.
Currently unavailable. The remote MCP endpoint is withdrawn pending an access-control review; use a local install.
Manual config (fallback)
If a client isn't auto-detected, run nucleus setup --dry-run to print every location nucleus looks in and the exact mcpServers block it would write, then paste that block into your client's config yourself. nucleus setup --create writes a Claude Code config even when nothing is detected. The full manual walkthrough lives in docs/QUICK_START.md, which ships with the package at /mcp_server_nucleus/docs/QUICK_START.md.
Nucleus finds your .brain automatically:
NUCLEUS_BRAIN_PATH environment variable (explicit).brain/ directory$HOME/.nucleus/brainThis lookup order applies to the running server. nucleus init instead seeds .brain/ in the directory you run it in (or the path argument) — set NUCLEUS_BRAIN_PATH afterwards if you want the server to find a brain placed somewhere else.
Nucleus has a full CLI alongside the MCP tools. Auto-detects TTY (table output) vs pipe (JSON).
# Memory
nucleus engram write my_key "insight here" --context Decision --intensity 7 nucleus engram search "compliance" nucleus engram query --context Strategy --limit 10
nucleus task list --status READY nucleus task add "Ship the feature" --priority 1
nucleus session save "Working on auth refactor" nucleus session resume
nucleus status --health nucleus sovereign
nucleus comply --jurisdiction eu-dora nucleus audit-report --format html -o report.html
nucleus chat
Pipe-friendly:
nucleus engram search "test" | jq '.key'
nucleus task list --format tsv | cut -f1,3
One-command configuration for regulatory frameworks:
nucleus comply --jurisdiction eu-dora # EU DORA
nucleus comply --jurisdiction sg-mas-trm # Singapore MAS TRM nucleus comply --jurisdiction us-soc2 # US SOC2
| Jurisdiction | Retention | HITL Ops | Kill Switch |
|---|---|---|---|
eu-dora | 7 years | 5 types | Required |
sg-mas-trm | 5 years | 5 types | Required |
us-soc2 | 1 year | 3 types | Optional |
global-default | 90 days | 2 types | Optional |
Nucleus collects anonymous, aggregate usage statistics (command name, duration, error type, versions, OS). No engram content, no file paths, no prompts, no API keys, no PII — ever.
nucleus config --no-telemetry
See TELEMETRY.md for details.
MIT © 2026 | hello@nucleusos.dev
Some deployments keep companion scripts in a repo next to this one. Point nucleus at them with either:
export NUCLEUS_SIBLING_REPOS="$HOME/repo-a:$HOME/repo-b" # os.pathsep-separated
printf '%s\n' "$HOME/repo-a" "$HOME/repo-b" > ~/.config/nucleus/siblings
With neither set, nucleus searches only its own repo — no assumptions are made about the host. A job whose script cannot be found fails with an error naming every location tried.
Nur der Anfang wird gezeigt; den Rest findest du im Repo