de en es ja ko pt ru zh

Local Worker MCP

Name: Local Worker MCP
Author: CaioAllgayer

by CaioAllgayer

Overview Schema Related Servers Score Discussions

Python

Local

本地工作器 MCP

本地MCP，客户端无关。前沿模型规划、委托和审查。本地执行繁重、机械且可验证的工作。

目标是减少付费AI的令牌消耗，而不将原始内容放入其上下文中。

Codex / Claude / Gemini / Grok
              │
              ▼
       Local Worker MCP
              │
       ┌──────┴───────────────┐
       ▼                      ▼
 Gemma 4 12B QAT          arquivos / PDF
 via Ollama               extração + evidências
       │
       ▼
 resultado compacto e verificável
       │
       ▼
 Frontier revisa

这不替代主AI，也不是模型路由器。这是真正的工作委托。

LOCAL DISPONÍVEL?          NÃO
      │                     │
      ▼                     ▼
   DELEGA                NÃO INSISTE
      │                     │
      ▼                     ▼
   COMPRIME            FRONTIER ASSUME
      │
      ▼
 FRONTIER REVISA

Gemma是一种优化。它绝不能成为单点故障。

原则

当任务机械、重复、可验证或上下文密集时委托：PDF、日志、CSV、代码、提取、分类、摘要。

保留在前沿：架构决策、安全、关键变更、主观判断、模糊需求。

如果本地工作器离线，前沿继续运行。MCP快速返回unavailable并推荐回退。客户端可以记录：

Worker local indisponível; executei diretamente.

不需要用户干预。

Related MCP server: ollama-handoff

要求

Python 3.10+
Ollama 在同一PC或局域网另一台PC上
一个本地模型（推荐：Gemma 4 12B QAT）

安装

git clone https://github.com/CaioAllgayer/Local-Worker-MCP.git
cd Local-Worker-MCP
python -m pip install -e ".[dev]"
copy .env.example .env

Ollama + Gemma

安装并启动Ollama。
下载模型。名称不是硬编码的——使用ollama list中的实际名称：

ollama list
ollama pull <nome-real-do-gemma>

如果LOCAL_LLM_MODEL为空，工作器尝试检测名称包含gemma的模型。否则，使用列表中的第一个模型。
测试端点：

curl http://127.0.0.1:11434/api/tags
local-worker status

启动MCP：

local-worker-mcp

同一PC

LOCAL_LLM_PROVIDER=ollama
LOCAL_LLM_BASE_URL=http://127.0.0.1:11434
LOCAL_LLM_MODEL=

local与lan通过主机名检测。127.0.0.1、localhost和::1为本地。

笔记本使用桌面

工作器不假设localhost。后端可以在局域网的另一台PC上。

在笔记本上：

LOCAL_LLM_PROVIDER=ollama
LOCAL_LLM_BASE_URL=http://192.168.x.x:11434

将192.168.x.x替换为桌面的实际IP（Windows上ipconfig，Linux上ip a）。项目中没有固定IP。

行为相同：快速失败、断路器、缓存、压缩。

在桌面上，Ollama需要接受局域网连接（变量OLLAMA_HOST=0.0.0.0，防火墙放行端口11434）。

OpenAI兼容

LM Studio、llama.cpp server、vLLM等：

LOCAL_LLM_PROVIDER=openai_compatible
LOCAL_LLM_BASE_URL=http://127.0.0.1:1234/v1
LOCAL_LLM_MODEL=...
LOCAL_LLM_API_KEY=

快速失败和断路器

默认值：

LOCAL_LLM_CONNECT_TIMEOUT_SECONDS=2
LOCAL_LLM_REQUEST_TIMEOUT_SECONDS=45
LOCAL_LLM_MAX_RETRIES=0
LOCAL_LLM_CIRCUIT_BREAKER_FAILURES=2
LOCAL_LLM_CIRCUIT_BREAKER_COOLDOWN_SECONDS=60

连接被拒绝不进入重试。N次失败后，电路断开，后续调用立即返回unavailable。冷却后，允许一次尝试。

{
  "status": "unavailable",
  "fallback_recommended": true,
  "reason": "Local LLM endpoint unreachable"
}

MCP工具

工具	功能
`local_status`	提供商、端点、本地/局域网、延迟、模型、断路器、缓存
`delegate_task`	通用任务 → 紧凑JSON
`delegate_batch`	并行独立任务（`MAX_PARALLEL_WORKERS=4`）
`delegate_file`	TXT、Markdown、CSV、JSON、代码、日志
`delegate_pdf`	按页提取、分块、层次综合、证据
`cache_stats`	大小、条目、命中、未命中、命中率、过期
`cache_cleanup`	立即GC（TTL → 不再使用 → LRU）
`cache_clear`	删除可丢弃条目

文件的原始内容不需要进入付费AI的上下文。工作器读取、压缩并返回可验证的证据（页、行、片段）。

安全

默认：SECURITY_MODE=READ_ONLY，ENABLE_SHELL=false。

SECURITY_MODE=READ_ONLY
ALLOWED_PATHS=C:\Projects,D:\Research
ENABLE_SHELL=false

READ_ONLY — 仅读取授权路径；写入和shell被阻止
WORKSPACE_WRITE — 在授权路径上读写；仅当ENABLE_SHELL=true时允许shell
FULL_LOCAL — 更宽松；仍阻止破坏性命令

路径遍历被阻止。rm、del、format等被拒绝。

缓存和日志

持久缓存位于~/.local-worker-mcp/cache，自动清理：

CACHE_TTL_DAYS=30
CACHE_MAX_SIZE_GB=10
CACHE_CLEANUP_THRESHOLD_PERCENT=90
CACHE_TARGET_USAGE_PERCENT=80
CACHE_CLEANUP_INTERVAL_HOURS=6

条目默认可丢弃。persistent=true保留重要工件。

日志轮转并过期：

LOG_RETENTION_DAYS=14
LOG_MAX_SIZE_MB=250

日志不保存文件的完整内容。

基准测试

local-worker benchmark arquivo.pdf

输出：

Arquivo: arquivo.pdf
Worker: gemma4:12b-qat
Backend: ollama
Endpoint: LAN/local
Tamanho: ...
Tokens originais estimados: ...
Tokens processados localmente: ...
Resultado para frontier: ...
Compressão: ...
Tempo: ...
Cache: HIT/MISS

Codex

~/.codex/config.toml或客户端的JSON：

{
  "mcpServers": {
    "local-worker": {
      "command": "local-worker-mcp",
      "env": {
        "LOCAL_LLM_PROVIDER": "ollama",
        "LOCAL_LLM_BASE_URL": "http://127.0.0.1:11434",
        "ALLOWED_PATHS": "C:\\Projects"
      }
    }
  }
}

参见examples/codex.json。

Claude Code

claude mcp add local-worker --scope user -- local-worker-mcp

或将examples/claude_code.json粘贴到~/.claude.json。

在项目的CLAUDE.md / AGENTS.md中，教导策略：

机械任务和大文件读取交给delegate_pdf / delegate_file / delegate_task。如果local_status或工具返回unavailable，直接执行并继续。

其他MCP客户端

任何stdio客户端都可以工作。通用示例在examples/generic.json：

{
  "mcpServers": {
    "local-worker": {
      "command": "local-worker-mcp",
      "env": {
        "LOCAL_LLM_PROVIDER": "ollama",
        "LOCAL_LLM_BASE_URL": "http://127.0.0.1:11434"
      }
    }
  }
}

示例

examples/pdf.md — 论文/长PDF
examples/code.md — 仓库初始读取
examples/logs.md — 错误提取

测试

python -m pip install -e ".[dev]"
pytest
ruff check src tests

测试套件不依赖真实的Ollama/Gemma。全部模拟。

此MVP不包括的内容

Windows GUI自动化、Playwright、复杂多智能体、向量RAG、仪表板、Kubernetes、ML路由器。

架构为下一阶段的delegate_repo、delegate_git、delegate_browser等留出空间。

许可证

MIT。

This server cannot be installed

license - permissive license

quality - not tested

maintenance

How are these scores calculated?

Maintenance

–Maintainers

–Response time

–Release cycle

–Releases (12mo)

Commit activity

Resources

GitHub Repository

Need Help?

Related Servers

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Servers

local-llm-delegation-mcp
mrrodriguez
A
license
A
quality
D
maintenance
Optimizes token costs by intelligently delegating low-complexity tasks to local LLMs via LiteLLM, enabling cost-effective development workflows.
Updated 3 months ago (2026-05-16 03:35 UTC)
3
1
MIT
ollama-handoff
AI & Machine Learning Developer Tools Autonomous Agents
Michael-WhiteCapData
A
license
A
quality
B
maintenance
Offloads cheap work from cloud LLM agents to a local Ollama model, reducing costs and keeping frontier models focused on complex tasks.
Updated 2 months ago (2026-06-23 02:07 UTC)
8
3
MIT
local-llm-mcp
Agent Orchestration AI & Machine Learning Text Summarization
HenryLinyy
A
license
B
quality
C
maintenance
Enables coding agents like Claude Code and Codex to offload boilerplate generation, summarization, and other bounded text tasks to local or cheap cloud LLMs, keeping the frontier agent in charge of judgment and code edits.
Updated 2 months ago (2026-07-01 00:26 UTC)
9
2
MIT
Relay
Agent Orchestration AI & Machine Learning
orhanucok
A
license
-
quality
C
maintenance
Enables AI to delegate boilerplate, drafts, tests, and refactors to free LLM providers, saving tokens and running tasks in parallel.
Updated 2 months ago (2026-06-30 20:55 UTC)
204
MIT

View all related MCP servers

Related MCP Connectors

ToolSnap MCP
Deterministic AI agent microtools, no accounts/API keys. fetch_extract: 98% token cut. 38 tools.
slipstream
Shared distillation cache for AI agents — every fetch ~73-89% fewer tokens via a shared cache.
netcafe-devkit
JSON/YAML, regex, diff, JWT, SQL dialects — the keyless millisecond ops an agent needs mid-task.

View all MCP Connectors

Latest Blog Posts

Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)
By Om-Shree-0709 on July 25, 2026.
mcp
Agent Identity
OAuth 2.1
Your AI Chatbot Just Exposed Your CEO's Salary to an Intern
By Om-Shree-0709 on July 2, 2026.
Agent Identity
MCP Security
OAuth Delegation
Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)
By Om-Shree-0709 on June 30, 2026.
Agentic Ai
Prompt Injection
WebAssembly

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/CaioAllgayer/Local-Worker-MCP'

If you have feedback or need assistance with the MCP directory API, please join our Discord server