能力标签

🔌 MCP 🤖 Agent 🔄 工作流 🐳 Docker 💻 CLI 🔗 REST API 🧬 Embedding 📚 RAG 🧠 Claude ✨ GPT

🔌

MCP工具

LoomCycle

基于 Go · 让 AI 助手直接操作你的系统与工具

英文名：loomcycle

⭐ 6 Stars 💻 Go 📄 Apache-2.0 🏷 AI 8.0分

8.0AI 综合评分

mcpagentic-aiai-agentsgolangllm

⬇ 下载源码 ZIP ⚙️ 配置说明

✦ AI Skill Hub 推荐

AI Skill Hub 强烈推荐：LoomCycle 是一款优质的MCP工具。AI 综合评分 8.0 分，在同类工具中表现稳健。如果你正在寻找可靠的MCP工具解决方案，这是一个值得深入了解的选择。

📚 深度解析

LoomCycle 是一款基于 MCP（Model Context Protocol）标准协议的 AI 工具扩展。MCP 协议由 Anthropic 开发并开源，旨在建立 AI 模型与外部工具之间的标准化通信接口，目前已被 Claude Desktop、Claude Code、Cursor 等主流 AI 工具采纳。

通过安装 LoomCycle，你的 AI 助手将获得额外的工具调用能力，可以用自然语言直接操控该工具的功能，无需学习复杂的命令行语法。MCP 工具的核心价值在于"一次配置，永久增强"——配置完成后，每次与 AI 对话时都可以无缝调用这些工具。

在技术实现上，MCP 工具通过标准的 JSON-RPC 协议与 AI 客户端通信，工具的功能以"工具列表"的形式暴露给 AI 模型，AI 可以按需调用。LoomCycle 提供了结构化的工具调用接口，使 AI 模型能够精确地理解和使用每个功能点，显著降低 AI 在工具使用上的错误率。

与传统的 API 集成相比，MCP 工具的优势在于无需编写代码——用户只需在配置文件中添加几行 JSON，即可让 AI 获得全新能力。AI Skill Hub 将 LoomCycle 评为 AI 评分 8.0 分，属于同类工具中的优质选择。

📋 工具概览

LoomCycle 是一款遵循 MCP（Model Context Protocol）标准协议的 AI 工具扩展。通过 MCP 协议，它可以让 Claude、Cursor 等主流 AI 客户端直接访问和操作外部工具、数据源和服务，实现 AI 能力的无缝扩展。无论是文件操作、数据库查询还是 API 调用，都可以通过自然语言在 AI 对话中直接触发，极大提升生产效率。

GitHub Stars

⭐ 6

开发语言

支持平台

Windows / macOS / Linux（跨平台）

维护状态

轻量级项目，按需更新

开源协议

Apache-2.0

AI 综合评分

8.0 分

工具类型

MCP工具

Forks

—

📖 中文文档

以下内容由 AI Skill Hub 根据项目信息自动整理，如需查看完整原始文档请访问底部「原始来源」。

📌 核心特色

通过标准 MCP 协议与 Claude、Cursor 等主流 AI 客户端深度集成
提供结构化工具调用接口，显著降低 AI 集成复杂度
支持 Claude Desktop 和 Claude Code 无缝接入，开箱即用
可与其他 MCP 工具组合叠加，构建完整 AI 工作站
轻量无侵入设计，不影响现有系统架构

🎯 主要使用场景

在 Claude Desktop 对话中直接调用本地工具，实现 AI 与系统的深度联动
通过自然语言驱动复杂的多步骤自动化任务，代替繁琐手动操作
将多个 MCP 工具组合使用，构建个人专属 AI 工作站

以下安装命令基于项目开发语言和类型自动生成，实际以官方 README 为准。

安装命令

# 方式一：通过 Claude Code CLI 一键安装
claude skill install https://github.com/denn-gubsky/loomcycle

# 方式二：手动配置 claude_desktop_config.json
{
  "mcpServers": {
    "loomcycle": {
      "command": "npx",
      "args": ["-y", "loomcycle"]
    }
  }
}

# 配置文件位置
# macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
# Windows: %APPDATA%/Claude/claude_desktop_config.json

📋 安装步骤说明

确认已安装 Node.js（v18 或以上版本）
打开 Claude Desktop 或 Claude Code 的 MCP 配置文件
按「交给 Agent 安装 → Claude Desktop」标签中的 JSON 配置填入 mcpServers 字段
保存配置文件并重启 Claude 客户端
重启后，在对话中即可使用本工具

以下用法示例由 AI Skill Hub 整理，涵盖最常见的使用场景。

常用命令 / 代码示例

# 安装后在 Claude 对话中直接使用
# 示例：
用户: 请帮我用 LoomCycle 执行以下任务...
Claude: [自动调用 LoomCycle MCP 工具处理请求]

# 查看可用工具列表
# 在 Claude 中输入："列出所有可用的 MCP 工具"

以下配置示例基于典型使用场景生成，具体参数请参照官方文档调整。

配置示例

// claude_desktop_config.json 配置示例
{
  "mcpServers": {
    "loomcycle": {
      "command": "npx",
      "args": ["-y", "loomcycle"],
      "env": {
        // "API_KEY": "your-api-key-here"
      }
    }
  }
}

// 保存后重启 Claude Desktop 生效

📑 README 深度解析真实文档完整度 44/100 查看 GitHub 原文 →

以下内容由系统直接从 GitHub README 解析整理，保留代码块、表格与列表结构。

简介

The agentic runtime, in a sidecar. One Go binary alongside your application — hardened agent loop, MCP on both sides, multi-replica HA. Apache-2.0.

🌐 <a href="https://loomcycle.dev">loomcycle.dev</a>  ·  📝 <a href="https://loomcycle.dev/blog/">Engineering blog</a>  ·  📐 <a href="https://github.com/denn-gubsky/loomcycle/blob/main/docs/ARCHITECTURE.md">Architecture</a>

---

🌳 Active development toward v1.0. The core primitives stabilised through v0.8 → v0.16 — multi-replica HA, the substrate Defs (Agent/Skill/MCPServer/Schedule/Webhook/MemoryBackend), A2A interoperability, inbound webhooks, pluggable memory + a memory layer, and a synthetic code provider. The feature set is complete; a hardening + QA pass remains before the v1.0 tag. OSS multi-tenant authorization (per-principal bearer tokens) is the headline of v1.1.0 — see "Planned" below. We welcome bug reports, security disclosures, feature contributions, downstream consumers, and forks. See CONTRIBUTING.md.

---

What's shipped

Capability	Released in
Six providers, native HTTP, no vendor SDK — Anthropic, OpenAI, DeepSeek, Gemini, Ollama cloud, Ollama local — behind one `Provider` interface with resolver-based routing per tier and effort	v0.4 → v0.8.x
Nineteen built-in tools including Claude Code parity (Read, Write, Edit, Grep, Glob, NotebookEdit), HTTP, WebFetch, WebSearch, Bash, Agent, Skill, Memory, Channel, AgentDef, SkillDef, Evaluation, Interruption, Context	v0.4 → v0.8.24
AgentDef + SkillDef + MCPServerDef substrate — content-addressed by SHA-256, runtime-mutable, push-at-boot from your container image; verify-or-fork across deployments	v0.8.5 / v0.8.22 / v0.9.x
Vector Memory with semantic search — `embed: true` on writes, `op: search` on reads. sqlite-vec or pgvector; Voyage / OpenAI / Gemini / nomic-embed on the embedding side	v0.9.0
MCP on both sides — loomcycle is both an MCP client (mounts external MCP servers as tools) and an MCP server (Claude Code and external orchestrators drive it via 21 meta-tools)	v0.8.15
OpenTelemetry across loop + providers + tools + MCP — no transcripts in spans	v0.10.0
Per-tenant fairness on the run-admitting semaphore (single-replica), cluster-wide (multi-replica)	v0.10.1 / v0.12.1
LLM Gateway + OpenAI-compatible shims — `POST /v1/_llm/chat`, `POST /v1/chat/completions`, `POST /v1/embeddings`. Drop loomcycle in front of any LangChain / LlamaIndex / n8n / RAG pipeline that speaks OpenAI's wire format	v0.11.0 → v0.11.4
n8n community package — `@loomcycle/n8n-nodes-loomcycle`. Five cluster sub-nodes, action nodes, two trigger nodes, six example workflows	v1.x (community)
Anthropic MAX subscription OAuth for dev workflow — reverse-engineered, dev-only, opt-in; see [`docs/PROVIDERS.md`](docs/PROVIDERS.md)	v0.11.10
Pause / Resume / Snapshot — runtime-wide quiesce + cross-version-portable JSON snapshot. In-place upgrades, snapshot-based replica handoff	v0.8.17 → v0.8.18
Multi-replica HA — Redis cancel pubsub, cross-replica run status, cluster-wide pause/resume + bus fanout, singleton sweepers, DB-backed session locks. Single binary scales from a cheap VPS to a multi-replica fleet	v0.12.0 → v0.12.5
UNIX-style trust model — operator config is the floor; callers narrow per-request but never widen. Bearer auth at the HTTP frontier; sandbox (Posture A) vs operator-trusted (Posture B) selected via env	v0.4 → ongoing
Embedded React Web UI at `/ui` — Library admin (agents / skills / MCP servers), Activity Monitor, Channels view, audit log, Schedules admin	v0.8.21 → v0.12.7
Multi-replica HA capstone — singleton sweepers, DB-backed session locks + hook registry, docs + hardening pass	v0.12.6
ScheduleDef substrate — operator-yaml `scheduled_runs:` templates + dynamic per-user forks with versioning + lineage. Sweeper fires due rows in parallel; 5-op tool (`create`/`fork`/`get`/`list`/`retire`) + 2 hook ops (`add_hook`/`remove_hook`); `on_complete` delivery via `channel.publish` / `memory.set` / `mcp.call`. Full 4-transport CRUD + `/ui/schedules` admin tab	v0.12.7
Per-run named credentials — `user_credentials: map<string,string>` on `POST /v1/runs` / gRPC / MCP / TS. `${run.credentials.<name>}` substitution in `mcp_servers.*.headers`. Back-compat: legacy `user_bearer` auto-promotes to `credentials.default`	v0.12.7
Bundled observability profiles — three `docker compose up` stacks: Grafana+Tempo+Prom+Loki (open-source), Honeycomb (SaaS), Datadog APM. `GET /metrics` Prometheus endpoint	v0.12.7
Mock LLM provider — cost-free 10K-agent stress harness. Variable latency + jitter + 429 / 5xx injection. Five model variants including `mock-mcp-caller` that exercises real MCP-tool dispatch end-to-end	v0.12.7
Curated MCP server recipe library — bundled JSON templates (GitHub, Slack, Tavily, …) + `$LOOMCYCLE_MCP_RECIPES_ROOT` filesystem overlay + a 7-verb `loomcycle mcp-registry` CLI that copy-pastes a recipe into your `mcp_servers:` yaml	v0.12.8
Claude Code repo import — `loomcycle import claude-code --from=./.claude/` walks agents / skills / mcp.json into loomcycle yaml (dry-run → review → `--write`); recipe-library rewrite, lossy-is-loud reporting. Paired with the `claude-code-plugin-loomcycle` Claude Code plugin (slash commands + skills + hooks over `loomcycle mcp`)	v0.12.8
Parent-context propagation — opaque `parent_context` {root_agent_run_id, function_key, tier_at_run} on a run is inherited by every Agent-tool sub-agent, persisted, and echoed on the agent-status / run-state / SSE surfaces — so a consumer can attribute a child sub-agent's usage to the user-initiated request	v0.12.8
Tool-use hooks — register HTTP webhooks against `(agent, tool, phase)` selectors via `POST /v1/hooks`. `pre` hooks rewrite a tool's input, deny it with a synthetic result, or (operator opt-in) widen hosts for one call; `post` hooks rewrite the result. Fail-open or fail-closed; narrows policy by default; DB-backed in cluster mode	v0.8.x → v0.12.5
A2A (Agent2Agent) interoperability — loomcycle as an A2A server (well-known AgentCard + REST/JSON-RPC/gRPC bindings) and client (call remote peers as `a2a__<peer>__<skill>` tools). Two substrate Defs, signed cards (JWS/JCS), multi-tenant routing, `INPUT_REQUIRED` ↔ Interruption resume. Off by default	v0.13.0
Input webhooks — `WebhookDef` substrate: external systems (GitHub/Stripe/Linear/CI/n8n) trigger agent runs or wake parked agents via signed `POST /v1/_webhooks/{name}`. HMAC-over-raw-body (verify-before-parse), strict JSONPath mapping, two-layer idempotency, per-Def rate limit, per-run credentials, `on_complete` hooks, admin triage. Off by default	v0.14.0
Memory ranking + pluggable backends — `Memory.search` hybrid ranker (semantic + recency weights) + search-time dedup; `MemoryBackend` interface with in-process default + `MemoryBackendDef` substrate + Mem9 REST backend (per-agent `memory_backend` routing, tenancy + RFC-F creds + graceful fallback); `loomcycle memory-eval` scoring harness. Opt-in, zero-regression default	v0.15.0
Memory layer (add / recall) — the `Memory` tool's optional LLM-extract paradigm: hand a backend conversation messages, it distils durable facts and answers natural-language recall (mem9 smart-mode). An optional capability alongside the frozen key/value `Backend`; the default in-process store fail-closes with `capability_unsupported`	v0.16.0
Synthetic `code-js` provider — `provider: code-js` runs operator JavaScript (goja) as a first-class agent for deterministic glue at zero token cost; stateless replay execution (resumable across restart/replica, deterministic by construction); default-deny tools, off by default	v0.16.0

Full per-version log: REVISIONS.md.

Security highlights

No vendor binary in the loop. Pure HTTP to provider APIs. No subprocess auth inheritance.
Default-deny everything. Every built-in tool is disabled until env-configured. Every agent gets zero tools until allowed_tools is set.
Two-layer policy + per-request narrowing. Operator floor in env; agent narrowing in yaml; caller narrowing per-run. Caller can never widen.
SSRF defence. Hostname allowlist + RFC1918/loopback/link-local IP block at the dial layer. Defeats DNS rebinding.
Constant-time bearer auth. sha256+CTC on both HTTP and gRPC.
Bash is restricted, not isolated. Run inside a container or VM if you need real isolation.

Full security model + the two-layer default-deny walkthrough: docs/TOOLS.md.

Install

Pick the path that fits. All four ship the same single static binary plus the v0.11.1 init / doctor first-run flow. `Context.help installation` covers each in detail.

```sh